Science Fair Guides · The Write-Up
How to write a
science fair report.
Judges read your report before they meet you, and most form a view on the first page. That is not unfair — it is what a clear report is for. This is what goes in each section, what students keep putting in the wrong one, and the rewrites that make a report read like science instead of homework. It takes three focused days and a willingness to delete sentences you were proud of.
← Back to the Science Fair Idea LabThe Structure
Why every paper is
shaped the same way.
The convention is called IMRaD: Introduction, Method, Results and Discussion. It looks like a rule teachers invented. It is the opposite — the shape that survived because it answers, in order, the only four questions a reader has. What did you want to know? What did you do? What happened? What does it mean?
It matters at a fair because nobody reads a paper front to back. A researcher reads the abstract, jumps to the figures, checks the method only if something looks off, then reads the discussion. A judge with forty reports behaves identically. IMRaD is a random-access structure, not a narrative one, so a reader can land anywhere and know what kind of claim they are looking at.
A fair report is IMRaD with pieces bolted on: title page, abstract, introduction with the literature review inside it, method, results, discussion, conclusion, references, appendices. Check your competition’s rules before formatting anything — abstract word limits are enforced, and the Society for Science ISEF rules change most years.
Order Of Writing
Write it out of order,
on purpose.
Everyone who starts with the introduction deletes it later. You cannot introduce a finding you have not articulated.
- 01Figures first — Built from your final dataset, before you write a sentence.
- 02Method — Write it with your notebook still open.
- 03Results — What each figure shows. No causal language.
- 04Discussion — What it means, how it compares, and your limitations.
- 05Introduction — Only now do you know what the report is about.
- 06Conclusion and references — A direct answer with a number. One consistent style.
- 07Abstract — Last. Always last.
- 08Read aloud, then checklist — Standing up, pencil in hand.
Section By Section
What belongs where.
Almost every problem in a student report is a placement problem. The content is there, in the section where the reader is not looking for it.
Abstract
~250 words. Written last.Belongs here
- +Context, then the question.
- +Method in one sentence, with the replicate count.
- +The headline result, with numbers.
- +One line on what it enables.
Ends up here wrongly
- −Promises. "This project will investigate" — it is finished.
- −Citations, or a compressed literature review.
- −No numbers: the fastest tell of a weak abstract.
Introduction
600–1000 words, literature included.Belongs here
- +The problem, narrowed to one unknown within two paragraphs.
- +What is known, with citations: your literature review.
- +The gap, and a hypothesis that could turn out wrong.
Ends up here wrongly
- −An encyclopedia entry on the general topic.
- −How excited you were. Save it for the interview.
- −Method detail, or results.
Materials and Method
500–1200 words. Past tense throughout.Belongs here
- +Materials with quantities, concentrations and models.
- +The procedure in the order it happened.
- +Replicates per condition, and how you chose that number.
- +Controls, instrument precision, software.
Ends up here wrongly
- −Results. "The solution turned blue" is a result.
- −A shopping list with no quantities.
- −Vagueness as formality: "the appropriate amount".
Results
Mostly figures. Short prose.Belongs here
- +Figures for trends, tables where exact values matter.
- +Means with a spread — SD, or median and range.
- +Test name, output, sample size.
- +Things that did not work, if they are data.
Ends up here wrongly
- −Interpretation. "Because" rarely belongs here.
- −Four hundred rows of raw data. That is an appendix.
- −Spreadsheet screenshots, or axes with no units.
Discussion
600–1000 words. Read hardest.Belongs here
- +What the results mean, in order of importance.
- +Comparison with published numbers, cited.
- +A mechanism, and your confidence in it.
- +Limitations, specifically, before anyone asks.
- +The next experiment, and why that one.
Ends up here wrongly
- −The results table restated as sentences.
- −New data appearing for the first time.
- −"Human error may have affected the results."
Conclusion
One paragraph. Sometimes three sentences.Belongs here
- +A direct answer to the introduction's question.
- +A number. Quantify the headline finding.
- +One line of honest scope.
Ends up here wrongly
- −Arguments appearing here for the first time.
- −A moral about saving the planet.
- −A second abstract.
References
Usually 8–20 entries.Belongs here
- +Every number, claim or method you did not invent.
- +One style — APA 7 unless told otherwise.
- +DOIs wherever they exist.
- +Software and datasets, cited like anything else.
Ends up here wrongly
- −Padding you never opened.
- −The news article instead of the study.
- −Bare URLs, or Wikipedia as a primary source.
The Stranger Test
Could someone you have
never met repeat this?
That is the whole standard for a method section. Not “did I describe what I did” but “could a stranger, with no access to me, get the same numbers?” The examples below use an invented study — how water temperature affects how fast an effervescent vitamin C tablet dissolves.
Fails the test
“We put a vitamin C tablet into water at different temperatures and timed how long it took to dissolve. We repeated the experiment a few times and recorded the results. Care was taken to keep everything else the same.”
A stranger can repeat nothing here. Which temperatures? How much water? How many is “a few”? What counted as dissolved?
Passes the test
“Each trial used 250 mL of deionised water in a 400 mL beaker, held within ±0.5 °C in a water bath and checked with a digital thermometer immediately before the tablet was added. Five temperatures were tested: 10, 20, 30, 40 and 50 °C. One 1000 mg effervescent ascorbic acid tablet, from a single batch, was dropped flat into the centre without stirring. Timing began on contact and stopped when no fragment was visible against a white tile, judged by the same observer throughout. Eight replicates ran at each temperature, in a randomised order so that drift in room conditions could not track the temperature sequence.”
Every guess a stranger would have to make is closed, including the two judges probe hardest: how the end-point was defined, and why the order was randomised.
To run the test: hand the method to someone not doing your project and ask them to describe back what they would do first, second, third. Every hesitation is a missing sentence. Do not explain out loud — write it down.
Results
Report what happened.
Say nothing about why.
In the results section you are a camera. You record; you do not explain. It is the hardest discipline in the draft, and the highest-value one:
“Mean dissolution time fell from 187 s (SD 14, n = 8) at 10 °C to 41 s (SD 6, n = 8) at 50 °C, a reduction of 78%, with intermediate means of 121 s, 84 s and 58 s. A one-way ANOVA across the five conditions gave F(4, 35) = 96.4, p < 0.001.”
No mechanism. No “as expected”. No “this proves”. Sentences that someone who disagreed with every idea you hold could still accept. Search your results section for those three phrases and move each into the discussion.
The separation is what makes the work checkable: if data and interpretation are welded into one sentence, anyone who doubts your explanation has to doubt your measurements too. It also changes the interview. When a judge offers a rival explanation and your results are clean, you can say: my data are consistent with that too, and here is what I would measure to tell them apart. If you have claimed the mechanism as a finding, the same question feels like an attack.
Discussion
Say the weakness out loud
before the judge does.
Judges are looking for the flaws in your design, and they will find them. The only thing you control is whether they find them in your report, written by you, or in the interview, extracted from you. Those score differently — not because judges admire modesty, but because naming a limitation proves you can look at your own design from the outside.
Write each one in four parts: name it, say which way it bends your result, estimate the size of the bend, and say what you would do instead.
Useless
“Human error may have affected the results. More trials could have been done and the equipment was not very accurate.”
Useful
“The end-point was judged by eye, and the last fragments were hardest to see at 10 °C, where the trial ran longest. That will have biased the 10 °C times upwards, probably by two or three per cent — smaller than the 78% effect, but not nothing. A conductivity probe logging at 1 Hz removes the judgement.”
The useful version sizes the flaw, so a reader can see it does not overturn the finding. Naming a limitation and leaving it unsized just makes them wonder how bad it is.
When it goes wrong
In an early version of my own microplastic write-up I reported removal percentages to two decimal places. The numbers came from counting particles down a microscope, by eye — a method with real counting uncertainty that I had described nowhere.
Someone asked how many particles I would have to miscount for the figure to change. I worked it out that evening: about three. Nothing about the experiment was wrong, but the write-up had overstated what it knew. The fix took an hour — report the counts, state the uncertainty, round to whole numbers. Never let your decimal places claim more than your instrument can deliver.
References
Citations, and never
formatting one by hand.
Cite the source of anything a sceptical reader could ask you to justify that you did not measure yourself: numbers, claims about what is known, methods you borrowed. Cite the study, not the news article about it — the article names the journal, so go and find the paper itself.
Never format references by hand: it is a pointless weekend and the output is always inconsistent. Install a reference manager on day one.
Free, open source, every operating system. The browser connector saves a paper with its metadata in one click, and the Word, Docs and LibreOffice plugins rebuild your reference list automatically when you change style.
The quote icon under any result gives a formatted citation and a BibTeX export. Use Cited by to walk forwards from a foundational paper.
A browser extension that finds legal, author-deposited free copies of paywalled papers.
Give it one paper and it maps the work around it — the fastest check on whether your gap is really a gap.
Set one style once. Include a DOI on every entry that has one — URLs rot, DOIs do not. Cite software and datasets too; if you analysed your data in JASP, that is a source like any other.
The Abstract
The hardest 250 words
in the document.
The abstract is the only part everyone reads and the only part some people read. It is also the part most students write in fifteen minutes at midnight, which is why it is so often the weakest page in a good project. Build it from five sentences: context, question, method with sample size, result with numbers, significance.
Weak — five sentences, zero numbers
“Chemistry is all around us, and dissolving happens every day. In this project I wanted to find out how temperature affects dissolving. I put vitamin C tablets into water at different temperatures and timed them. I found the hotter the water, the faster the tablet dissolved, which was what I expected. This shows temperature is important for chemical reactions.”
Strong — four sentences, twenty-two numbers
“Effervescent tablets are a common delivery form for vitamin C, but packaging specifies no water temperature, even though dissolution rate decides how long a user waits. I measured dissolution time for 1000 mg tablets in 250 mL of deionised water at five temperatures (10, 20, 30, 40 and 50 °C, ±0.5 °C), eight replicates each, order randomised, with the end-point defined as no solid visible against a white tile. Mean time fell from 187 s (SD 14) at 10 °C to 41 s (SD 6) at 50 °C, a 78% reduction, and a one-way ANOVA gave F(4, 35) = 96.4, p < 0.001. Warming cold tap water by 20 °C more than halves the wait — an instruction packaging does not give.”
Same project. The weak version reports that an experiment happened; the strong one reports a finding. Count the numbers in your own abstract, then check every one appears identically in the results — that mismatch is the most common error I see, and it costs you credibility instantly.
Reading It Aloud
Your eye reads what you meant.
Your mouth reads what you wrote.
Silent reading of your own draft is nearly useless after the second pass, because your brain supplies the sentence you intended. Your mouth can only follow the words on the page. Three signals, each with a fix. Out of breath: too long, split it. Stumbling: the grammar is tangled, rewrite as subject, verb, object. Bored by your own voice: the passage carries no information, cut it.
Unsayable — 39 words
“It was found that in the trials which were carried out at the higher temperatures that were tested, the times which were recorded for the tablets to become fully dissolved were lower than those recorded at the lower temperatures.”
Sayable — 18 words
“Tablets dissolved faster in warmer water: 187 s at 10 °C, falling to 41 s at 50 °C.”
Shorter, and it carries data the long version did not.
Stand up and read at speaking pace. Hold a pencil, mark every hesitation, and fix nothing while you read — stopping to edit destroys the flow you are testing for. Repair the marks in a second pass. If you hate your own voice, use text-to-speech; it will not smooth over a broken sentence out of loyalty to you. Hemingway Editor is free in the browser and flags the sentences you would have stumbled on. One last trick: read your reference list from the bottom up. Reversing it breaks the skim that hides a missing year.
Before You Submit
The pre-submission
checklist.
Work through this the day before, not the hour before submission.
- □Every figure has axis labels, units, a caption and a sample size.
- □Every number in the abstract appears identically in the results.
- □The introduction's hypothesis is the one the discussion answers.
- □The method names the replicates per condition.
- □No result sits in the method; no interpretation in the results.
- □Limitations are specific, directional and sized.
- □Every citation has a matching reference entry, and the reverse.
- □Word and abstract limits match this year's rules.
- □Required forms are signed and attached.
- □Read aloud, start to finish, with a pencil in your hand.
- □The raw data lives somewhere that is not only your laptop.
Common Questions
What students ask
about the write-up.
How long should a science fair report be?
For most fairs the body runs to eight to fifteen pages excluding appendices, but length is not what is marked. A workable split is 600–1000 words of introduction, 500–1200 of method, a results section that is mostly figures, and 600–1000 of discussion. A six-page report a stranger could repeat beats twenty pages of padding.
What is the difference between the results and the discussion?
Results are what happened; the discussion is what it means. The results section reports measurements and figures with no causal language, so "it dissolved faster because the molecules move faster" belongs in the discussion. Keeping them apart lets a reader accept your data while disagreeing with your explanation — which is what a good judge tries to do.
How do I write a science fair abstract?
Write it last, from five sentences: context, question, method with sample size, result with numbers, and what it means. Most fairs cap it at around 250 words, so every sentence has to carry a fact. Count the numbers as a diagnostic — if there are none, you have summarised your intentions rather than your findings.
How do I write limitations without hurting my chances?
Naming a limitation yourself is what makes you look like a scientist rather than a student defending a claim, and judges find the weaknesses either way. Use four parts: name it, say which way it bends your result, estimate the size of the bend, and say what you would do instead. A sized limitation earns more than a hidden one costs.
How many references does a science fair report need?
Eight to twenty is normal, but the count matters less than whether you read them. Cite the source of every number, claim or method you did not invent, and cite the original study rather than a news article about it. Never list a paper you could not summarise if a judge asked.
Where To Go Next
Keep going.
The free tool that builds a project around your interests and equipment — or stress-tests the idea you have.
Which test to run, how many replicates you need, and what a p-value does not tell a judge.
Your report is read; your board is scanned in eight seconds. Different writing.
The paperwork that disqualifies more good projects than bad science.
The report is not paperwork you do after the science. Writing it is when you find out what you actually think — every year I meet students who find a hole in their reasoning on the day they write the discussion. That is the process working.
— Fionn
Google Science Fair Global Grand Prize, 2019
Science Fair Guides · Idea Lab · fionnferreira.com