For standard SUS coding, subtract 1 from odd-numbered responses and subtract even-numbered responses from 5. Sum the ten contributions and multiply by 2.5. Interpret the 0–100 result as perceived usability, with the study context and an explicit reference—not as a percentage of satisfaction or task success.
Use SUS to measure perceived usability after use
The System Usability Scale is John Brooke's ten-item questionnaire for an overall subjective assessment of a system's usability. It produces a score from 0 to 100. That score is neither the percentage of tasks completed nor the percentage of people satisfied. Keep those measures separate even when they appear in the same research report.
Use the established instrument rather than writing ten agreeable statements and calling the result SUS. Brooke's original chapter, linked in the references, contains the questionnaire and scoring instructions. Administer it after people have used the system and before a debrief that might change how they describe their experience.
This guide follows an invented study of a library room-booking prototype. Participants attempt to find a suitable room, book it, and change the booking. Five complete SUS responses and one incomplete response are used only to demonstrate the analysis. The six responses make the scoring and missing-data rules easy to inspect; choose a benchmarking sample around the precision your decision needs.
Define the experience the score is supposed to describe
A participant cannot rate a system they have not meaningfully encountered. A picture of a home page might support a first-impression study, but it does not provide the same experience as completing a booking and handling a change. Decide what people will use, for how long, and under what conditions before collecting the questionnaire.
For the fictional study, the target is the room-booking prototype after the same three tasks. Record the version and any moderator assistance. Someone who completed the work independently may have experienced a different system burden from someone whose errors were repaired by the moderator. Keep that difference visible rather than treating the questionnaire as a substitute for observing the tasks.
If you repeat the measurement after a redesign, preserve the intended user group, task scope, device conditions, and administration as far as the comparison requires. A change from new users on phones to experienced staff on desktops can change the score even if the design has not improved.
| Field | Fictional study decision | Reason to record it |
|---|---|---|
| System | Room-booking prototype, revision B | Identify exactly what was evaluated |
| Participants | Adults who have not used this booking flow | Avoid mixing first use with expert familiarity |
| Exposure | Find, book, and reschedule a room | Give each respondent a defined experience |
| Administration | Questionnaire before discussing the session | Avoid a debrief supplying the interpretation |
| Additional evidence | Task outcomes, assistance, and observed problems | Explain behavior that a global score cannot locate |
| Missing data | Report complete responses and incomplete forms separately | Keep the score denominator visible |
Score one response all the way through
For the standard item order and 1–5 response coding, subtract 1 from each odd-numbered response. Subtract each even-numbered response from 5. Each contribution then runs from 0 to 4. Add the ten contributions and multiply the total by 2.5. This reverses the negatively worded items so that higher contributions consistently point toward a more favorable assessment.
Participant P01 has the fictional responses 4, 2, 4, 2, 3, 3, 4, 2, 4, 2. The ten contributions sum to 28, so P01's score is 28 × 2.5 = 70. The calculation is shown below using item numbers rather than replacing the questionnaire's wording.
A quick check catches common spreadsheet errors: the most favorable pattern—5 on odd items and 1 on even items—must produce 100; the opposite must produce 0. All responses at 3 produce 50. Merely adding the raw answers fails because agreement has different meanings across the alternating items.
| Item and raw response | Contribution calculation | Contribution |
|---|---|---|
| 1: 4 | 4 − 1 | 3 |
| 2: 2 | 5 − 2 | 3 |
| 3: 4 | 4 − 1 | 3 |
| 4: 2 | 5 − 2 | 3 |
| 5: 3 | 3 − 1 | 2 |
| 6: 3 | 5 − 3 | 2 |
| 7: 4 | 4 − 1 | 3 |
| 8: 2 | 5 − 2 | 3 |
| 9: 4 | 4 − 1 | 3 |
| 10: 2 | 5 − 2 | 3 |
| Total | Sum contributions | 28 |
| SUS score | 28 × 2.5 | 70 |
Keep the individual scores and the denominator
Calculate a complete score for each respondent before summarizing the group. In this fictional example, the five complete scores are 70, 55, 60, 75, and 90. Their sum is 350 and their mean is 350/5 = 70. The median is also 70, but the observed range of 55 to 90 shows that people did not all report the same experience.
P06 skipped an item. That is a missing response, not an observed neutral answer. This example uses a predeclared complete-response analysis, so P06 contributes no total and the report states five complete forms out of six collected. A different missing-data method would need to be specified and justified; silently inserting a 3 would change the data.
For a benchmark intended to generalize, plan the sampling and uncertainty analysis before collecting responses. These five invented scores illustrate arithmetic only. Their mean cannot establish a population percentile or demonstrate that a redesign has improved usability. A larger count alone would also not correct a biased recruitment process.
| Respondent | SUS score | Treatment in this example |
|---|---|---|
| P01 | 70 | Complete; calculation shown above |
| P02 | 55 | Complete |
| P03 | 60 | Complete |
| P04 | 75 | Complete |
| P05 | 90 | Complete |
| P06 | Not calculated | One missing response; excluded from complete-form mean |
| Mean of complete forms | 70 | 350 divided by 5; do not divide by 6 |
Interpret a score against an explicit reference
MeasuringU's published interpretation guide places 68 around the center of its reference distribution and distinguishes raw SUS scores from percentile ranks. Its example maps a score of 75 to approximately the 73rd percentile in that reference. The mapping is empirical; 75 is not automatically the 75th percentile just because both numbers use a 0–100 range.
A reference distribution helps orient a discussion, but it is not a universal release threshold. A broad average does not establish that a booking flow is accessible, that it avoids harmful errors, or that your intended users can complete a critical task. Choose the comparison relevant to the decision and state where its interpretation comes from.
If a previous study had a mean of 65 and a later study had a mean of 70, the descriptive difference would be 5 SUS points. Do not call that a five-percentage-point increase in satisfaction. Whether it reflects a meaningful, repeatable change depends on the samples, study conditions, variability, and the comparison design—not just the subtraction.
| Number | What it describes | What it does not establish |
|---|---|---|
| SUS 70 | A transformed questionnaire score | 70% of users satisfied or 70% of tasks successful |
| A percentile from a named reference | Position relative to that reference distribution | A property inherent in the raw score |
| 5-point change between study means | A descriptive difference in SUS units | A causal redesign effect without a suitable comparison |
Connect the score to observed problems
Suppose P05, who scored 90, still failed to confirm a rescheduled booking. Their favorable overall impression does not erase the failed task. Conversely, a participant might complete every task but find the process exhausting. Perceived usability and observed effectiveness can diverge; the disagreement is a reason to examine the session.
The hypothetical research team therefore keeps a separate outcome log. It records whether the new booking was actually confirmed and whether the person understood the pending state. The next design decision is to clarify that state and retest it. Changing a label solely because one questionnaire item scored lower would jump from a broad attitude measure to a diagnosis the questionnaire does not provide.
Use interviews and task observations to locate the problem. Use the repeated questionnaire, administered consistently, to track the broader experience. Neither replaces the other, and a high overall score should not be used to waive a known failure in the main user journey.
| Evidence in the fictional study | Interpretation | Next action |
|---|---|---|
| P05 reports SUS 90 but leaves a change pending | Positive impression coexists with a failed task | Inspect the confirmation sequence and retest it |
| Several people ask whether a change is final | The status wording may not match their understanding | Test a clearer pending-versus-confirmed presentation |
| Mean SUS is 70 across five complete forms | A descriptive summary of this small sample | Report scope and variability; do not announce a universal pass |
Write a report another researcher can check
A usable result includes the system, participants, exposure, administration, scoring, missingness, and limits. The numerical summary should be short enough to understand while leaving the underlying records available for checking. Preserve item-level responses so a scoring mistake can be corrected without repeating the study.
For the fictional case, a defensible summary is: 'After three tasks on revision B of the room-booking prototype, five of six participants supplied complete SUS forms. The complete-form mean and median were 70, with scores from 55 to 90. One incomplete form was excluded under the stated rule. These descriptive results from a small first-use sample do not establish population performance. The rescheduling confirmation problem remains a priority despite favorable ratings from some participants.'
For the next round, record what changed and why before comparing scores. If the participants, tasks, assistance, or questionnaire version also changed, state those differences. The purpose of measuring is to improve a decision about the experience, not to produce a more flattering number.
- Name the evaluated version, intended users, tasks, device conditions, and timing of the questionnaire.
- Link the instrument source and record any language version or adaptation used.
- Report the number collected, number scored, missing-data rule, and descriptive distribution.
- Name the benchmark or uncertainty method if used; distinguish it from the raw score.
- Pair the result with observed task outcomes and a concrete next research or design decision.