In July 1945, at a test site in the New Mexico desert, the world's first nuclear bomb was detonated. Enrico Fermi was watching from about 10 miles away. As the shockwave reached him, he tore a sheet of paper into small pieces and dropped them from chest height. He watched how far the blast scattered them, did some quick arithmetic in his head, and announced his estimate of the bomb's yield: about 10 kilotons.
The actual yield, measured precisely by instruments, was 18.6 kilotons. Fermi was within a factor of 2 - using torn paper.
That instinct for arriving at a defensible number from almost nothing is what Fermi problems are named after.
What Is a Fermi Problem?
A Fermi problem is a question designed to be solved not through memory or lookup, but through estimation and structured reasoning.
The questions are usually about quantities too large or obscure to know off the top of your head:
- How many piano tuners are there in Chicago?
- How many gas stations are in the United States?
- How many golf balls fit inside a Boeing 747?
The point isn't to get the exact answer. The point is to arrive at the right order of magnitude - to know whether the answer is in the thousands, the millions, or the billions.
The Piano Tuner Problem, Solved
This is the most famous Fermi problem, and it's worth working through properly rather than just referencing it.
How many piano tuners are there in Chicago?
Start with what you roughly know. Chicago's population is about 3 million people. An average household has about 2.5 people, so there are roughly 1.2 million households.
What fraction of households have a piano? Pianos are moderately expensive and not all that common in most homes. Perhaps 1 in 20 households - that's about 60,000 pianos in Chicago.
How often does a piano need tuning? Active pianos typically get tuned once or twice a year. Call it 60,000 tunings per year.
How many tunings can a piano tuner do in a day? A tuning takes roughly 2 hours including travel time, so maybe 4 per day. Working about 250 days a year, that's 1,000 tunings per tuner per year.
So: 60,000 tunings ÷ 1,000 per tuner = about 60 piano tuners in Chicago.
The actual figure, based on records at the time Fermi posed this question, was around 50 to 80. The estimate lands right in the middle.
Why the Technique Works
This seems like it shouldn't work. You're making a chain of rough guesses, each potentially off by quite a bit. How can the final answer be anywhere close?
The answer is that errors tend to cancel. When you overestimate one quantity, you often underestimate another. A chain of estimates that are each within a factor of 2 or 3 will usually produce a final result within a factor of 5 or 10 of the true answer - which is exactly the level of precision Fermi problems are designed to test.
This also explains why breaking a problem into more steps often helps rather than hurts. More steps means more opportunities for errors to average out.
Where Fermi Problems Show Up
The technique is used seriously in physics, engineering, finance, and medicine. Doctors estimate drug dosages based on body weight assumptions. Engineers sanity-check designs with back-of-the-envelope calculations before committing to detailed models. Investors estimate market sizes before deciding whether a company is worth pursuing.
The same reasoning applies to understanding the world more generally. Knowing whether a statistic is plausible - whether a claimed number is in the right order of magnitude - is a genuinely useful skill when reading news, evaluating claims, or making decisions with incomplete information.
For a worked example of what this looks like on a real question, see the chicken population estimation, which walks through exactly this kind of breakdown step by step.
The Key Mindset
The thing that separates good Fermi estimators from poor ones isn't knowledge - it's the willingness to commit to a number.
Most people, when faced with "how many piano tuners are in Chicago?", say "I have no idea." That's a refusal to engage with the problem, not an honest answer. You almost certainly know something relevant. You know Chicago is a large city. You know pianos exist and need tuning. You know roughly how many hours are in a workday.
From there, you can build.
The discipline is committing to your best estimate of each piece and letting the multiplication do the rest. Being wrong by a factor of 2 on three inputs still gets you within an order of magnitude of the answer. That's usually good enough to be useful, and it's infinitely more valuable than declining to try.
For a broader look at how this reasoning style translates into a game format - and why getting the scale right matters more than precision - see What Is an Estimation Game? and How Big Is a Billion, Really?