NASA-TLX calculator: raw and weighted task load index

The NASA Task Load Index was developed by Sandra Hart and Lowell Staveland at NASA Ames in the 1980s and has since been used in several thousand studies. It measures perceived workload on six dimensions, each rated 0–100 on a 21-point line:

The original procedure adds a weighting step: the respondent is shown all 15 pairs of dimensions and picks, for each pair, the one that contributed more to the workload of this task. A dimension can win at most five comparisons, and the weights always sum to 15. The idea is that time pressure matters more in air-traffic control than in proofreading, and the respondent should say so rather than the analyst assuming it.

Raw TLX, and why almost everybody uses it

Hart’s own 2006 review of 20 years of TLX use found that most researchers had quietly dropped the weighting step and averaged the six ratings instead, and that this “raw TLX” behaves about as well — sometimes more sensitively, sometimes less, rarely differently enough to change a conclusion. That is why this page shows both and reports the gap between them: if the weighting moves your score by a point, it did not matter; if it moves it by ten, your respondents are telling you something about which dimension dominates.

Performance is the reversed one

Five dimensions run low → high in the direction of more workload. Performance runs Perfect → Failure, so on the original form a high number means the person did badly, which counts as high workload. Survey tools routinely record it the friendly way round (good = 100), and the resulting scores are wrong in a way that is hard to spot because they still look plausible. The checkbox on this page flips it.

TLX is a well-validated instrument that is easy to misuse. These are the failure modes that matter.

• There is no universal “good” score. This page deliberately does not print a grade, because a global benchmark for TLX does not exist in the way it does for the System Usability Scale. Published scores vary enormously by domain and by task — a demanding cockpit task and a demanding spreadsheet task are not on a shared scale. Grier’s 2015 meta-analysis of around a thousand tasks is the place to look for distributions by task type, and its central message is that comparisons are only meaningful within a task type. Any tool that hands you an A–F grade for a TLX score has invented it.
• It is only useful as a comparison. TLX earns its keep comparing two designs, two tasks or the same task before and after a change — with the same participants, the same task wording and the same moment of asking. A single number for a single condition is almost uninterpretable.
• It measures perceived workload, not performance or usability. A task can feel easy and be done wrong, or feel hard and be done perfectly. TLX does not replace task success, time on task or the System Usability Scale; it explains part of why they came out as they did.
• The Performance subscale confuses respondents as well as analysts. Some people rate how well they think they did; some rate how satisfied they are; some read the reversed anchors the wrong way. If Performance is your biggest driver, look at the raw answers before believing it.
• Averaging the six dimensions hides the finding. The score is a sum of things that are not commensurable. Two conditions with the same mean can differ completely in profile, and the profile is what you can act on.
• Weights are about the task, not the person. The pairwise comparisons must be answered for the specific task just performed. Reusing one respondent’s weights across several tasks, or averaging weights and applying them to everyone, quietly turns the weighted score back into a raw one with extra steps.
• Repeated measures break the confidence interval. The interval on this page treats each row as an independent respondent. If the same people rated several tasks, you need a within-subjects analysis; the interval shown here will be too narrow.
• Retrospective ratings drift. Asked ten minutes later, people rate the peak of the task rather than its average. Ask immediately, and ask the same way every time.
• Small samples are normal and still small. Eight participants is a common TLX study and a wide interval. Report the interval and the profile; do not report a two-point difference between conditions as a result.

Used as a paired comparison with a stable protocol, TLX is one of the most useful instruments in the toolbox — it tells you which kind of hard a design is. Used as an absolute score with a made-up grade, it is decoration.

Most TLX literature is about aircraft, operating theatres and driving simulators, but the instrument transfers to software work with two adjustments.

• Pick a task, not a screen. “Invite a teammate and give them edit access”, not “the settings page”. Workload is a property of what someone is trying to do.
• Expect physical demand to be near zero and keep it anyway. It is a useful floor: a respondent rating physical demand at 60 for a web task has misread the scale, and the row is worth checking.
• First-run tasks are the interesting ones. Mental demand and frustration during a first attempt are what onboarding is supposed to reduce, and they move faster than satisfaction scores do. Pair TLX with time to value and the funnel to see whether a lighter-feeling flow actually converts.
• Compare before and after the same way. Same task text, same point in the session, same participants if you can get them. Then a five-point drop in mental demand is evidence rather than noise.
• Watch the tooltip and tour copy. Temporal demand and frustration climb when in-product guidance moves faster than people read. The tooltip reading time calculator and the tooltip copy checker deal with exactly that.

1. Hart S. G., Staveland L. E. “Development of NASA-TLX (Task Load Index): Results of empirical and theoretical research.” In Hancock P. A., Meshkati N. (eds), Human Mental Workload, North-Holland, 1988 — the instrument, the six dimensions and the pairwise weighting procedure.
2. Hart S. G. “NASA-Task Load Index (NASA-TLX); 20 Years Later.” Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 2006 — how TLX is used in practice, and the standing of raw TLX.
3. Grier R. A. “How High is High? A Meta-Analysis of NASA-TLX Global Workload Scores.” Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 2015 — distributions of global TLX by task type; the reason this page does not print a universal grade.
4. NASA Human Systems Integration Division, NASA Task Load Index: NASA states that TLX is open source, available for use worldwide, and that permission is not required either to use it or to modify it. This page implements the published scoring and is not affiliated with or endorsed by NASA.