Does N-Back Training Improve IQ Reasoning-Test Scores?

Does getting better at N-back mean your reasoning improved?

No. Better performance on the practiced N-back task is a direct practice gain. Improvement on an untrained but similar N-back task is near transfer. Improvement on a distinct reasoning measure is farther transfer, and evidence does not establish a dependable gain in general IQ-style reasoning. Someone may become more effective at tracking information in N-back without gaining an advantage on a task that asks them to infer a new pattern.

N-back asks a person to monitor a changing stream and decide whether the current item matches one presented a fixed number of steps earlier. Repeating this operation can improve performance through task familiarity and strategy. An untrained N-back changes the items or presentation but retains much of the operation, making it a closer test. A matrix-reasoning task instead asks someone to infer a relation among visual elements and use it to complete a pattern.

Fluid reasoning means solving problems that require working out relationships in unfamiliar material rather than mainly recalling learned facts. An IQ-style test samples performance through particular tasks, instructions, timing, and scoring rules; it samples rather than observes all reasoning capacity. A score increase on one measure supports only a limited statement about performance on that measure under those conditions. To argue that N-back caused the change, researchers must consider what would have happened without practice and whether a comparison group shows the same change.

The key comparison is the difference between a training group and a credible active-control group on the same untrained reasoning measure before and after. An active control provides another structured activity to help account for time, expectations, and repeat testing. A trainee's increase alone cannot separate training effects from test familiarity or ordinary variation. Keep results on the practiced task and an untrained N-back separate from the reasoning outcome.

The N-back-specific meta-analysis, “Working Memory Training Revisited: A Multi-Level Meta-Analysis of N-Back Training Studies,” combined 203 effects from 33 randomized controlled trials . It found a medium-sized effect on untrained N-back tasks, while effects on other working-memory tasks, fluid intelligence, and cognitive-control tasks were similarly very small. A broader review, “Working Memory Training Does Not Improve Performance on Measures of Intelligence or Other Measures of ‘Far Transfer,’” examined 87 publications and 145 comparisons of working-memory programs. It found no convincing reliable far-transfer improvement against treated controls. Because this review was not limited to N-back, it provides context rather than a direct N-back estimate.

Ask exactly which untrained outcome improved, and relative to which comparison group. Better performance on the exercise shows learning of that exercise; improvement on a similar memory task suggests near transfer. A controlled advantage on an unfamiliar reasoning measure would be relevant evidence of farther transfer, but one measure alone would not establish a general change in intelligence.

Sources: Working Memory Training Revisited: A Multi-Level Meta-Analysis of N-Back Training Studies; Working Memory Training Does Not Improve Performance on Measures of Intelligence or Other Measures of ‘Far Transfer’: Evidence From a Meta-Analytic Review

Why could N-back practice plausibly affect reasoning, and what did the first positive study show?

N-back practice could plausibly affect some reasoning performance because both can involve keeping information active while updating it, but overlap in mental demands is a hypothesis for transfer, not proof that the abilities or tests are interchangeable. Fluid reasoning means solving new problems without relying mainly on learned facts. An intelligence test samples performance through particular tasks and conditions; a change on one measure does not by itself establish a broad change in reasoning. In an N-back task, a person follows a stream of items and, whenever a new one appears, decides whether it matches the item presented a set number of steps earlier. For example, imagine hearing a sequence of animal names and being asked to notice whether the current name is the same as the one heard two turns ago. The person must retain the relevant recent items, replace older ones as the stream continues, and make a comparison at the right point. This illustrates the updating demand without reproducing a test item. Some reasoning tasks also require retaining intermediate information while comparing relationships or revising a possible solution. That resemblance makes transfer worth testing. It does not show that practice with one stream improves performance on a different task, nor that any change would reflect a single underlying capacity. The influential 2008 study, “Improving fluid intelligence with training on working memory,” reported transfer from a demanding working-memory task to measures of fluid intelligence. Its abstract describes adaptive dual N-back training and says the reported gain depended on the amount of training: greater training was associated with greater improvement on the reasoning measures. The record identifies the study as a 2008 report in the Proceedings of the National Academy of Sciences. This was a positive finding, so the careful conclusion cannot be that N-back training is incapable of affecting any reasoning-test result. The finding instead supports a narrower proposition: under the protocol and outcome conditions studied, the researchers observed improvement beyond the training task and reported a dose relationship. That result remains one study’s finding, not a conversion rule for an individual. A dose relationship within a study does not tell a reader how many training sessions would produce a particular score change, whether the same pattern would appear with another protocol, or whether the change would last. The PubMed abstract does not give enough detail to establish those broader claims, so they should not be inferred from its summary. Nor does the label “fluid intelligence” make its outcome equivalent to every reasoning test, a normed IQ score, or a browser quiz estimate. The proposed mechanism and the observed result should therefore be kept separate. Updating may explain why researchers thought N-back could transfer; the 2008 result provides evidence that transfer was observed in that study. Neither establishes that updating caused the result, that the effect generalizes reliably, or that every person should expect improvement. The finding is important because it keeps a protocol-specific benefit possible. To decide whether it is dependable, the next question is how later studies compared N-back with credible control activities and whether their untrained reasoning results agreed.

Sources: Improving fluid intelligence with training on working memory

Why do meta-analyses reach different estimates?

The 2015 and 2017 N-back meta-analyses can both be accurately described without forcing them into a single yes-or-no verdict. The first pooled studies measuring fluid intelligence and reported a small positive average effect. The second separated improvement on an untrained N-back task from transfer to other working-memory tasks and fluid-intelligence measures; its clearest effect was on the untrained N-back outcome, while effects in the other domains were very small. A broader 2016 review of working-memory training found no convincing reliable far-transfer benefit against treated controls. “Fluid intelligence” means performance on measures designed to assess reasoning through novel problems rather than recall of learned facts. A pooled effect averages contrasts across included studies; it does not promise an individual gain or translate into a fixed number of IQ points. “Improving Fluid Intelligence with Training on Working Memory: A Meta-Analysis” focused on N-back studies with a fluid-intelligence outcome, a control group, and healthy participants aged 18 to 50. The authors report that 20 studies met those criteria and found a small, statistically significant positive effect on fluid-intelligence measures. This supports possible transfer under those study conditions, but not a dependable individual gain or lasting everyday benefit. “Working Memory Training Revisited: A Multi-Level Meta-Analysis of N-Back Training Studies” analyzed healthy adults trained with N-back, drawing 203 effect sizes from 33 randomized controlled trials. Importantly, it treated untrained N-back tasks as a separate transfer category instead of combining them with other working-memory outcomes. The authors found a medium-sized effect on untrained N-back tasks, but very small effects for other working-memory tasks, fluid intelligence, and cognitive control. Another N-back result is not a reasoning result. The syntheses also differ in analytic choices, including which outcomes they emphasize and how multiple effects are handled. The 2017 analysis used a multilevel method that included effects from relevant tasks in original studies and reported domains separately. “Does performance carry to another N-back?” is a different question from “Does it carry to fluid reasoning?” “Working Memory Training Does Not Improve Performance on Measures of Intelligence or Other Measures of ‘Far Transfer’” adds an important boundary, but it is not an N-back-only review. It covered 87 publications and 145 comparisons of working-memory programs with pretest, posttest, and control groups. It found reliable immediate gains on verbal and visuospatial working-memory measures, but no convincing evidence of reliable improvement on more distant outcomes such as ability, reading, and arithmetic when training was compared with a treated control. A treated control performs another activity, helping account for participation, time, and expectations. Because that review combines different programs, it cannot settle the N-back-specific estimate. It cautions that pre/post change or comparison with a passive group is weaker evidence than advantage over another credible activity. Small pooled estimates also do not prove every protocol has zero effect; they summarize included studies and their limits. For the reader’s decision, the best synthesis is narrower than either “N-back raises IQ” or “transfer is impossible.” Practice improves the trained task, and the 2017 review found a medium-sized effect on untrained N-back tasks. Evidence for a dependable gain on distinct reasoning measures is less consistent: one N-back-specific synthesis reported a small positive fluid-intelligence effect, another found very small effects in that domain, and a broader review found no convincing far-transfer benefit against treated controls. Any improvement claim should name the reasoning measure and comparison. It should not turn a pooled research effect into a personal IQ-point forecast.

Sources: Improving Fluid Intelligence with Training on Working Memory: A Meta-Analysis; Working Memory Training Revisited: A Multi-Level Meta-Analysis of N-Back Training Studies; Working Memory Training Does Not Improve Performance on Measures of Intelligence or Other Measures of ‘Far Transfer’: Evidence From a Meta-Analytic Review

What does an active-control replication actually test?

An active-control study asks a sharper question than “Did trainees score higher afterward?” It asks whether N-back training produced a larger change than another credible activity, when both groups also repeated the outcome tests. That comparison helps separate a training-specific effect from changes that could follow time, attention, engagement, expectations, or familiarity with testing. A no-contact group can show what happened without scheduled practice, but it does not control those experiences as closely as an activity matched for contact and effort. For judging transfer, the key result is therefore the difference between groups on an untrained reasoning measure, not the N-back group’s pre-to-post change by itself. In “No Evidence of Intelligence Improvement After Working Memory Training: A Randomized, Placebo-Controlled Study,” young adults received 20 practice sessions in either adaptive dual N-back or adaptive visual search; a third group received no contact. The adaptive visual-search activity served as an active placebo control, giving researchers a way to compare two practiced activities as well as either one against no scheduled practice. Participants completed pretest, midtest, and posttest sessions with multiple measures covering fluid intelligence, multitasking, working-memory capacity, crystallized intelligence, and perceptual speed. The abstract reports that performance improved on both practiced activities, but the researchers found no positive transfer to any of the cognitive-ability tests. The study therefore shows that improvement during training can coexist with no detected advantage on the broader measures it tested. It does not show that every N-back protocol must have exactly zero effect on every reasoning outcome. This distinction matters when comparing the study with “Improving Fluid Intelligence with Training on Working Memory,” the favorable 2008 report. That paper reported transfer from demanding working-memory training to fluid-intelligence measures and said the gain varied with training time. The reports do not answer identical questions merely because both included N-back practice and reasoning outcomes. The later study used an adaptive visual-search activity as an active comparison and measured several cognitive abilities at three test points; the 2008 abstract describes transfer under its own training conditions. The accessible records support this contrast in reported design and result, but they do not justify filling gaps in one study’s control details from assumptions. A positive finding under one protocol remains relevant evidence; a later study with a more informative comparison tests whether a gain can be distinguished from the effects of another engaging activity. An active control also does not settle every design concern. Activities can differ in expectations or in how closely they match the mental demands of training, and a study can fail to detect a small effect. So “no positive transfer” is a statement about the tested comparisons and measures, not proof of impossibility. Nor should comparison quality be confused with sample size: adding participants cannot by itself make an unbalanced comparison answer the causal question, while a strong comparison still leaves uncertainty about effects too small or specific to detect. The practical reading rule is simple: look for a between-group advantage on an untrained reasoning test, with a credible activity control and repeated measures. Treat a trainee’s score increase alone as an observed change, not evidence that N-back caused broader reasoning improvement.

Sources: No Evidence of Intelligence Improvement After Working Memory Training: A Randomized, Placebo-Controlled Study; Improving fluid intelligence with training on working memory

What does the 2022 mediation result add—and what does it leave unresolved?

The 2022 study “Near transfer to an unrelated N-back task mediates the effect of N-back working memory training on matrix reasoning” offers an explanation for mixed transfer findings. Across three randomized controlled trials involving 460 participants, researchers reported that improvement on untrained N-back tasks statistically mediated transfer to Matrix Reasoning in all three trials. Yet the overall intervention effect on Matrix Reasoning was inconsistent: trials two and three did not show an intervention effect. The finding suggests a pathway worth testing; it does not establish a dependable average reasoning gain from N-back training.

Mediation means a statistical model finds that a measured change helps account for a relationship between an intervention and an outcome. Here the proposed chain was N-back training, better performance on an untrained N-back task, and then Matrix Reasoning performance. “Untrained” matters: the transfer task was not the practiced N-back itself, but it remained a similar memory task. This is evidence about near transfer possibly relating to a farther outcome, not a direct demonstration that practice broadly raises intelligence.

An indirect pathway and an overall intervention effect answer different questions. A mediation model can find relationships among training, an intermediate measure, and an outcome even when the average outcome difference between training and control groups is not statistically detectable. The pattern may help explain variation across studies, as the Nature Human Behaviour abstract proposes. But mediation alone does not prove that the intermediate gain caused the reasoning outcome, or turn an inconsistent intervention result into a reliable benefit. A plausible sequence can motivate a mechanism hypothesis; it cannot substitute for a replicated effect on the outcome readers care about.

The paper also reports a significant intervention effect on a working-memory composite in trial three, unlike Matrix Reasoning. Its abstract says untrained N-back performance mediated transfer to both outcomes. This gives researchers a testable account of why transfer might depend on what participants learn. Results for an untrained N-back task, a working-memory composite, and matrix reasoning should not be collapsed into a broad label such as “higher IQ.”

The authors propose that transfer to untrained N-back tasks may gate further transfer, including to Matrix Reasoning. That is a candidate model, not a settled mechanism. The repeated mediation pattern supports examining near-transfer gains as a possible condition for farther transfer; absent intervention effects in two trials limit the claim that training reliably improves reasoning. Neither finding cancels the other. Together they leave room for protocol-specific effects while withholding a general promise.

For someone deciding whether to train, this study supports curiosity, not expectation. N-back may improve similar untrained-task performance, and a particular protocol may be associated with a reasoning gain under some conditions. The study does not establish that an individual should expect such a gain, that it will persist, or that it represents changed general intelligence. Stronger evidence would replicate an intervention effect on untrained reasoning in controlled trials, report near-transfer and reasoning outcomes separately, and test the proposed pathway in advance. Until then, mediation is an explanation to investigate, not a settled answer.

Sources: Near transfer to an unrelated N-back task mediates the effect of N-back working memory training on matrix reasoning; Near transfer to an unrelated N-back task mediates the effect of N-back working memory training on matrix reasoning

Why does improvement on the training task fail to settle the IQ question?

Because success on the practiced exercise and success on an untrained reasoning test are different outcomes. Repeated exposure can improve task-specific strategies and familiarity; that is a real performance change, but it does not by itself identify a broader reasoning change. To argue that training caused transfer, researchers need a distinct outcome and a comparison group that helps estimate what would have happened without N-back practice.

A useful way to keep the claims in order is an outcome ladder: the practiced N-back task; a different, untrained N-back task; another memory task; matrix reasoning; and broader measures of cognitive ability. This is not a ranking of results. It specifies what a transfer claim covers. Improvement on a second N-back may extend beyond one exact exercise while leaving open whether learning carries over to a different kind of problem.

“Putting brain training to the test” followed 11,430 participants in a six-week online study. They trained several times a week on multiple cognitive tasks involving reasoning, memory, planning, visuospatial skills, and attention. The study reported improvement on every trained task, but no evidence of transfer to untrained tasks, even when cognitively close. This illustrates why a training score alone cannot answer the IQ question. It was not an N-back-only experiment, however, and cannot settle whether a particular N-back protocol transfers to reasoning.

The N-back-specific multilevel meta-analysis of 33 randomized trials makes a related distinction: its abstract reports a medium effect on untrained N-back tasks, alongside very small effects on fluid intelligence and other assessed transfer domains. This pattern fits a gradient in which learning is clearest near the trained operation and less evident on more distant outcomes. It does not prove transfer impossible or predict one person's score. Evidence at one rung cannot simply be carried up the ladder as if the outcomes were interchangeable.

A before-and-after reasoning score also has multiple possible explanations. On a repeated form, a person may remember instructions, grow more comfortable with the format, or adopt a strategy suited to those items. A different form reduces direct item recall but may share the same format or demands. The increase may be real; it alone does not identify its cause. Raw performance, such as items answered correctly, describes that administration. A normed interpretation requires a suitable reference group and conversion; neither follows from a raw change by itself.

When evaluating a claim, ask: Was the outcome practiced or untrained? How far did it differ from N-back? Was the same form repeated? Did the training group improve more than a credible control group on that outcome? Similar improvement in both groups leaves retesting or other shared influences plausible. A controlled N-back advantage on matrix reasoning supports a claim about performance on that measure under those study conditions. It does not by itself establish a general change in intelligence.

Sources: Putting brain training to the test; Working Memory Training Revisited: A Multi-Level Meta-Analysis of N-Back Training Studies

Can a browser IQ-style score tell you whether N-back worked?

No single browser score can establish that N-back training worked. Test IQ Free’s 50-question result covers pattern, number, and spatial reasoning, but its estimated range is a provisional educational estimate from an unvalidated conversion with no validated adult reference population. The product page says it is not a validated IQ measurement or a confidence interval, and it is not a normed comparison with adults. A different range after practice therefore cannot show that N-back caused a change in general intelligence. The distinction is between what a person did on the questions and what a score is taken to mean. The answers and domain feedback describe observed performance on this particular set of reasoning items under the conditions in which it was completed. A norm-referenced score, by contrast, interprets performance against results from a defined comparison group. Test IQ Free does not have a validated adult reference population for its conversion, so its displayed estimate cannot supply that population comparison. The familiar IQ label does not create evidence for a clinical or norm-referenced interpretation. A before-and-after change also leaves several explanations open. The person may have become more familiar with the item format, approached the questions with different effort, or completed the test in different circumstances. Performance can vary across occasions. Without a validated score scale and a comparison group, the size or direction of a change in the displayed band cannot isolate an N-back effect. That is why training studies use a different comparison. In “Working Memory Training Does Not Improve Performance on Measures of Intelligence or Other Measures of ‘Far Transfer’: Evidence From a Meta-Analytic Review,” the authors synthesized 87 publications with 145 experimental comparisons. They reported reliable immediate gains on verbal and visuospatial working-memory measures, but no convincing evidence of reliable far-transfer improvement against treated controls on measures including nonverbal ability, verbal ability, reading, and arithmetic. The review covered working-memory programs broadly, not N-back alone; it helps explain why gains relative to an active activity are more informative than a trainee’s own score increase. It does not validate or directly evaluate Test IQ Free’s conversion. For that reason, Test IQ Free’s result is not a standardized baseline for testing whether N-back changed reasoning. A useful training experiment would compare groups under similar conditions and examine the same untrained reasoning measure in both, rather than treating one person’s change as causal evidence. For low-stakes curiosity, it can still be an exploratory reasoning exercise. The optional report adds explanation and a practice plan, not accuracy or validity for the estimate. For consequential education, clinical, or other decisions, use an appropriately validated professional assessment instead of treating this browser result as decisive.

Sources: Test IQ Free; Working Memory Training Does Not Improve Performance on Measures of Intelligence or Other Measures of ‘Far Transfer’: Evidence From a Meta-Analytic Review

What is a responsible next step if you want to improve?

Start by naming the outcome you want. If you want to answer N-back sequences more accurately or quickly, practice N-back and judge progress on comparable N-back tasks. If your goal is broader reasoning, include varied, unfamiliar problems and review why each solution works. That gives practice with different operations; it does not guarantee that an IQ score or general intelligence will rise. Keep the goal and evidence aligned. A better score on the exercise you practiced shows improved performance on that exercise. To ask whether a separate reasoning skill changed, measure it directly under similar conditions; do not treat one before-and-after difference as proof that training caused it. Note the task, conditions, and observed performance. Worked answers can help you understand an operation instead of memorizing an item. For a low-stakes check of reasoning performance, the Test IQ Free test page offers a free 50-question assessment across pattern, number, and spatial reasoning. Its estimated range is provisional, produced by an unvalidated conversion with no validated adult reference population. It is not a normed comparison or evidence that training changed general intelligence. Use the questions and worked answers as exploratory practice. The responsible conclusion is narrow: N-back is practice for N-back; varied reasoning practice can target reasoning tasks, while a dependable IQ gain has not been established.

Sources: Test IQ Free

Questions readers ask