“How many golf balls would fit inside a school bus?” was the most-cited Google brainteaser of the 2000s and the question Laszlo Bock specifically named when he publicly retired the brainteaser format in 2013. For about a decade, the question was a rite of passage for any candidate interviewing in tech — a Fermi-estimation problem that tested whether the candidate could break a giant unfamiliar quantity into ratios, decompose those ratios into component estimates, and combine them coherently.
The pure question is dead in tech. Its descendants — Fermi estimation in product manager interviews, market sizing in consulting interviews, capacity planning in staff-engineer system design rounds — are alive. Knowing how to do the calculation cleanly is still useful, even if the literal “how many golf balls” question is no longer asked.
The full estimation
The standard back-of-envelope works in three layers: estimate the volume of the bus interior, estimate the volume of a golf ball, divide, then apply a packing efficiency correction.
Bus interior volume. A typical school bus is about 30 feet long, 8 feet wide, and 6 feet tall internally. Volume is 30 × 8 × 6 = 1,440 cubic feet.
Golf ball volume. A standard golf ball has diameter 1.68 inches, so radius ≈ 0.84 inches ≈ 0.07 feet. Volume of a sphere is (4/3) × π × r³ = (4/3) × π × (0.07)³ ≈ 1.4 × 10⁻³ cubic feet. So roughly 700 golf balls per cubic foot if we ignored packing.
Naive count. Bus volume ÷ golf ball volume = 1,440 / 0.0014 ≈ 1,000,000. So about a million golf balls if we packed them perfectly.
Packing efficiency correction. Spheres in a random packing fill about 64% of the volume; in optimal hexagonal close packing about 74%. Using a 64% factor: 1,000,000 × 0.64 = 640,000 golf balls.
Subtract seats and structure. About 25% of the bus interior is taken up by seats, the driver area, the wheel wells, and other obstructions. So multiply by 0.75: 640,000 × 0.75 = 480,000.
Final answer: approximately 500,000 golf balls. Anywhere in the range of 300,000 to 1,000,000 is a defensible answer. The point is not the exact number; the point is the structure of the reasoning.
Why this is “Fermi”
Enrico Fermi was a 20th-century physicist who used to challenge his students at the University of Chicago with order-of-magnitude estimation problems. His most famous one, “how many piano tuners are in Chicago”, became the namesake for this style of reasoning. Fermi estimation is the practice of producing a defensible answer to a quantitative question without doing any actual research — just decomposing into ratios you can estimate from general knowledge.
The Google interview question is a near-perfect Fermi problem. It is too big to be intuited directly. It decomposes naturally into multiplicative components (volume of bus × density of golf balls × correction factors). And the answer is robust to substantial errors in any individual component, because errors in different components partially cancel.
What the question was supposed to test
Google’s stated theory in the 2000s was that Fermi estimation tests:
- Decomposition. Can the candidate break a large problem into multiplicative components?
- Defensible numbers. Can they attach a defensible estimate to each component without panicking about precision?
- Sanity checking. Does the final answer make sense at the order-of-magnitude level?
- Composure. Can they reason aloud through an unfamiliar problem in 5–10 minutes without freezing?
For roles that involve a lot of uncertain quantitative reasoning — product management, business analyst, finance — these are real skills. Google’s mistake (the eventual conclusion of their internal data) was applying the same test to software engineering candidates, where the underlying skills are different.
The 2013 retirement
Laszlo Bock’s 2013 New York Times interview specifically named this question as an example of a brainteaser that “doesn’t predict anything”. Google’s internal People Analytics team had correlated brainteaser performance with on-the-job software engineering performance and found near-zero correlation. The retirement was public, named, and effective: by 2015 the question was extinct at Google, and by 2018 it was mostly extinct industry-wide.
Where it survives
- Product manager interviews. “How many ride-share rides happen in Tokyo on a Friday night?” tests the same skill, in a domain where the answer is plausibly relevant to the actual job (sizing markets and demand).
- Management consulting interviews. McKinsey, BCG, and Bain still use market-sizing questions extensively. “How many tennis balls are sold in Germany annually?” or “What is the total revenue of the US dental market?” are McKinsey-format Fermi problems.
- Staff engineer capacity planning. “Estimate the storage cost of running this at 100x scale” is a Fermi problem with engineering-relevant numbers. Senior+ system design rounds use this format regularly.
- Finance and trading interviews. “How many trades does the NYSE process per second on average?” Again, the answer is somewhat relevant to a trader’s worldview.
The pattern: Fermi estimation survives in roles where the underlying skill (quick, defensible reasoning under uncertainty) is genuinely job-relevant, and dies in roles where it is not.
The signal in a strong answer
The interviewer is watching for four things:
- Clarification first. “What kind of school bus — 40-passenger or smaller? Is the bus stripped of seats? Are we including the engine compartment?” Asking these before computing signals professional maturity.
- Structured decomposition. Volume → unit volume → packing efficiency → obstructions. The chain of multiplicative factors should be explicit.
- Numbers with reasoning. Not “a school bus is 1,500 cubic feet” but “a school bus is roughly 30 feet by 8 feet by 6 feet, so about 1,500 cubic feet”. The numbers are easier to defend when their derivation is visible.
- Order-of-magnitude reasonableness. The candidate notices that “100 golf balls” is wrong and “100 billion” is wrong, then converges on a range like “hundreds of thousands to a million”.
What not to do: confidently produce a single number without showing work. Confidently produce numbers that are off by orders of magnitude. Refuse to engage with the problem because “I do not know how big a school bus is”.
Variations interviewers ask
- “Pieces of luggage in a 747.” Same shape, different vehicle. Standard variation.
- “Cigarettes that fit in a phone booth.” Older variation; nobody asks about phone booths in 2026.
- “How many cars cross the Brooklyn Bridge in a day?” Combines geometry (lane width, traffic density) with time (peak vs off-peak).
- “How many gallons of paint to paint the Eiffel Tower?” Surface area × paint coverage rate.
- “How many barbers are there in the United States?” Population × frequency × productivity. Different decomposition (population-based, not volume-based).
The population-based and volume-based decompositions are the two main flavors of Fermi problems. Most candidates can handle one of the two; strong candidates can recognize which decomposition is appropriate for a given question.
Is it asked in 2026?
The literal question “how many golf balls fit in a school bus” is essentially never asked in 2026. Asking it earnestly would be a strong signal that the interviewer has not updated their process in 15 years. The Fermi-estimation skill, however, is alive in PM, consulting, finance, and senior system-design rounds, and preparing for those formats means being able to do calculations of exactly this shape under verbal pressure.
Frequently Asked Questions
What’s the “right” answer?
Anywhere from 300,000 to 1,000,000 is defensible with reasonable assumptions. The exact number is less important than the structure of the reasoning.
Did Google really retire this question?
Yes. Laszlo Bock specifically named it in his 2013 New York Times interview as a “complete waste of time”. By 2015 it was extinct in serious Google interviews.
Should I prepare Fermi estimation in 2026?
For tech engineering roles, no. For PM, consulting, finance, or senior+ system design rounds, yes. The skill is real even though the literal question is dead.
What is the cleanest way to estimate a school bus’s volume?
Length × width × height. Don’t try to model the curved roof or the exact shape; treat it as a rectangular prism with a 25% obstruction correction.
Why does the packing factor matter?
Spheres can’t fill space without gaps. Random packing of spheres fills about 64% of the volume; even optimal hexagonal close packing fills only 74%. Forgetting the packing correction overestimates the answer by about 50%.