How Many Questions Should an Exam Have?
Plan exam length using learning outcomes, question types, timing and a worked 60-minute paper example.
There is no universal number of questions an exam should have. The right count depends on what must be covered, the types of response, the time available and the evidence needed for the decision. A 60-minute paper with 30 short multiple-choice items is a different assessment from a 60-minute paper with two extended responses, even though both could be well designed.
Start with learning outcomes and a test blueprint, then estimate actual response time. The University of Waterloo’s guide to preparing tests and exams recommends checking whether a knowledgeable assistant can complete the draft comfortably within the permitted time. Treat that as a pilot, not as a fixed multiplier for students.
Use a time budget, not a magic item count
For a 60-minute classroom exam, first reserve time for instructions, reading and checking. Suppose the team allocates 5 minutes for these tasks, leaving 55 minutes of response time. A draft might use:
| Section | Count | Working estimate | Time |
|---|---|---|---|
| Multiple choice | 15 | 1.5 minutes each | 22.5 minutes |
| Short answer | 5 | 4 minutes each | 20 minutes |
| Extended response | 1 | 12 minutes | 12 minutes |
| Response time | 54.5 minutes | ||
This totals 21 questions and 59.5 minutes including the 5-minute allowance. The numbers are an illustrative planning model, not a rule. If the extended response includes a long source text or diagrams, the reading estimate must increase. If a question requires showing calculations, its time estimate should include writing, not just solving mentally.
Then check content coverage
A paper can fit the clock yet miss important outcomes. Make a small matrix: topics down the rows; recall, application and analysis across the columns. Place each planned item in one primary cell. If the test is supposed to represent a term of learning but half its marks depend on a single week, revise the mix.
Some outcomes need more than one observation. A single item may be distorted by wording, context or a momentary mistake. ETS’s guide to test reliability discusses how the number of questions relates to score consistency, but adding weak or redundant items is not a substitute for sampling the intended content well.
Count marks and cognitive work as well as questions
Twenty one-mark recall questions and one twenty-mark essay should not be described as “mostly recall” merely because 20 of 21 prompts are recall. The marks are split 50/50. Review three totals separately: number of questions, marks and estimated time by topic and thinking level. Those views can reveal different imbalances.
Five checks before finalising the count
- Coverage: Does each priority outcome have enough evidence?
- Time: Can someone complete the full paper, including reading and checking, with a reasonable margin?
- Marks: Are marks roughly proportionate to the evidence and effort requested?
- Accessibility: Does the layout add avoidable reading or navigation time?
- Decision: Is the paper long enough for its stakes, without asking students to repeat the same task?
What if the paper is too long?
Do not cut questions at random. First remove duplicates, overlong background material and prompts that assess low-priority facts. Then rebalance the blueprint. A short paper can still be weak if it no longer samples the curriculum. Conversely, a longer paper can be weak if it mainly repeats one easy skill.
For a reusable exam workflow, keep timing, marks, type and topic with each question. When building a paper, these fields make the estimate visible before publication. Our guide to balanced exam patterns shows how to arrange the selected items into coherent sections.
Practical answer: draft from the blueprint, calculate a time budget, pilot the whole paper and adjust until both coverage and completion time make sense. The resulting count may be 12, 21 or 40; the justification matters more than the number.
Two papers can have the same length and different quality
Imagine two 30-question tests. Paper A has 30 one-step recall items from two recent lessons. Paper B has 20 short items that sample the course and 10 application items tied to priority outcomes. Paper B is not automatically better, but its count represents more varied evidence. A question count is a workload measure, not a validity measure. The blueprint must explain why each section exists.
Likewise, two ten-question papers may have very different completion times. Ten lengthy data-response tasks can take longer than 30 factual items. Estimate reading and response time per task, then validate the estimates with a realistic trial. If a student must choose questions, pilot the longest reasonable choice combination, not just the shortest one.