How to write good multiple-choice questions: rules, examples and common mistakes
Ten research-based rules for multiple-choice questions with before-and-after examples from maths, science and geography, and mistakes that help guessing.
In short: A good multiple-choice question has one clearly best answer, wrong answers built from real student mistakes, and a stem that asks a complete question. Leave out "all of the above", avoid negative wording, keep the options alike, give no clues in the wording and spread the correct answer across A, B, C and D. Most of these rules come from a review of item-writing guidelines by Haladyna, Downing and Rodriguez (2002).
Where do these rules come from?
Most of the rules below come from "A review of multiple-choice item-writing guidelines for classroom assessment" by Haladyna, Downing and Rodriguez (2002). The authors checked 31 guidelines for classroom tests against 27 measurement textbooks and 27 studies and reviews published since 1990.
On the number of options, a meta-analysis of 80 years of research by Rodriguez (2005) concluded that three good options are usually enough. An older 1989 list said the opposite, so online rule lists that mix the two contradict themselves.
Ten rules for better multiple-choice questions
1. Write one clearly best answer
Exactly one option should be correct beyond doubt. If a colleague could argue for a second one, rewrite the question. Ask for "a square number" while offering both 16 and 64, and two answers are right.
2. Build wrong answers from real student mistakes
A wrong answer, or distractor, only works if an unprepared student might pick it. The 2002 review recommends plausible distractors built from typical student errors. The best source is your last test: the wrong answers your class actually chose.
3. Put the whole question in the stem
Cover the options: if a student who knows the topic could still answer, the stem is complete. Leave out details that have nothing to do with the answer.
4. Leave out "all of the above" and "none of the above"
The 2002 review says to avoid "all of the above" and to use "none of the above" carefully. Cynthia Brame's guide for the Vanderbilt Center for Teaching advises against both. A student who spots two correct options can pick "all of the above" without knowing the third.
5. Word the stem positively, or make the NOT impossible to miss
Under time pressure, "Which of these is not a mammal?" is easily read without the "not". Ask the positive question where you can. If the negative is the point, capitalise it and set it in bold: "Which of these is NOT a mammal?"
6. Keep the options alike
Options should be about the same length, the same kind of thing (all cities, all years) and in the same grammatical form. A visibly longer or more precise correct answer is a well-known giveaway.
7. Give no clues in the wording
Look for hints that need no knowledge of the topic:
- "always" or "never", which make an option easy to reject,
- a word from the stem repeated in one option, such as "condenses" in the question and "condensation" among the options,
- grammar that fits only one option, such as a stem ending in "is called an" when only "adjective" starts with a vowel,
- two options that mean the same thing, or plainly absurd ones.
8. Ask about one idea per question
Test one piece of content and one kind of thinking per question. Avoid combinations like "1 and 3 are correct", and keep questions independent so that one does not give away another.
9. Vary the position of the correct answer
If the correct answer is often B, students notice. Spread the correct answers evenly over A, B, C and D, without long runs of the same letter.
10. Test understanding, not only recall
A question copied from the textbook tests whether a student recognises a sentence. The 2002 review suggests new material for higher-level thinking: paraphrase, and give a new example, calculation or situation.
Before and after: three examples
Geography: wrong answers nobody would pick
Before: Which city is the capital of Australia? A) Canberra, B) Paris, C) Tokyo, D) Africa.
Paris and Tokyo belong to other countries and Africa is not a city, so even a student who knows nothing about Australia gets it right.
After: the same question with A) Sydney, B) Melbourne, C) Perth, D) Canberra. Correct answer: D. All four are Australian cities, and Sydney catches the most common mistake.
Science: a clue in the stem
Before: Water vapour condenses on a cold window. What is this process called? A) Condensation, B) Evaporation, C) Melting, D) Freezing.
"Condenses" points straight at "Condensation", and the question only asks for a name.
After: On a cold morning, drops of water appear on the inside of a car window. Where does the water come from?
- A) Rain that soaked in through the glass
- B) Water released by the cold glass itself
- C) Water vapour in the air inside the car
- D) Cold air from outside that turned into water
Correct answer: C. The student applies condensation to a new situation, and each wrong answer is something a student could genuinely believe.
Maths: recognising a formula versus using it
Before: (a + b)² = A) a² + 2ab + b², B) 2ab, C) ab, D) a + b.
The stem is not a question, and only one option looks like the formula from the textbook.
After: Which expression is equal to (x + 3)²?
- A) x² + 9
- B) x² + 6x + 9
- C) x² + 3x + 9
- D) 2x + 6
Correct answer: B. Each wrong answer is a typical mistake: leaving out the middle term (A), not doubling it (C), or doubling instead of squaring (D).
How Grade AI fits in
Grade AI (grade-ai.app) is a phone and web app for teachers that prints multiple-choice tests on paper, one answer sheet per student, and grades the filled-in sheets with the phone camera.
Four options, one correct answer. Every Grade AI question has options A to D and one key; the editor will not continue until all four are filled in. That fits the research as long as all three wrong answers are plausible.
AI drafts are checked, but not for everything. When Grade AI drafts questions, the AI is told to write exactly one correct answer and three plausible wrong ones. The app then throws out drafts with "all of the above" or "none of the above", two identical options or the correct answer written in the question. A wrong option with the same value as the correct one (4/6 next to 2/3) gets a different number, or the draft is dropped. The correct answers of AI drafts are spread evenly across A to D. The app cannot catch two answers that are both correct in meaning, a wrong answer key or a reworded duplicate. Questions you write by hand keep their letters as you set them. Check every question against these rules before printing. Details in AI test maker for teachers and AI test questions from your own notes.
The results show which wrong answer the class chose. After scanning, the answer distribution shows for every question how many students picked A, B, C and D. A distractor nobody picks is doing no work. When at least 40 percent of the students who wrote the test, and at least three, chose the same wrong answer on the hardest question, the test Overview says in one sentence that either the question is misleading or the class learned it wrong. The "Reteach tomorrow" card in Statistics shows the question with the lowest success rate, if below 60 percent, with the wrong answer picked most often. More in Which questions did the class get wrong?
Sources
- Haladyna, T. M., Downing, S. M. and Rodriguez, M. C. (2002). A review of multiple-choice item-writing guidelines for classroom assessment. Applied Measurement in Education, 15(3), 309-333. doi.org/10.1207/S15324818AME1503_5
- Rodriguez, M. C. (2005). Three options are optimal for multiple-choice items: A meta-analysis of 80 years of research. Educational Measurement: Issues and Practice, 24(2), 3-13. doi.org/10.1111/j.1745-3992.2005.00006.x
- Brame, C. J. Writing good multiple choice test questions. Vanderbilt Center for Teaching, adapted by the University of Wisconsin-Madison. wisc.pb.unizin.org
Frequently asked questions
How many options should a multiple-choice question have?
Three good options are usually enough. A meta-analysis of 80 years of research (Rodriguez, 2005) found that with three options you can ask more questions in the same time without harming test quality. What matters is that every wrong option is plausible. Grade AI always uses four options, A to D.
Is "all of the above" a bad option?
Usually, yes. The 2002 review of item-writing guidelines recommends avoiding it. A student who sees that two options are true can pick it without knowing the third.
How do I stop students from guessing?
You cannot stop it completely: with four options, a blind guess is right one time in four. Convincing wrong answers, no clues in the wording and more questions make guessing pay less. In Grade AI every question is worth one point, with no deduction for wrong answers.
Can AI write good multiple-choice questions?
AI can write a useful first draft, but check it for a second correct answer, a wrong key and distractors nobody would pick. In Grade AI every AI question is marked "Suggested by AI", and you can edit or delete it before printing.
Try it on your next test.
Set up a class, print the sheets, see how long it takes. Scanning happens on the phone. Building and printing work in the browser too.