How Accurate Is Punxsutawney Phil? We Scored Every Recent Prediction
NOAA puts Phil at 35% accurate over the last 20 years — 17th out of 19 US groundhogs. The reason isn't bad luck. It's that he calls “six more weeks of winter” in roughly 85% of years, and late winter has stopped cooperating.
A groundhog is not a forecaster, and the record proves it. NOAA, which grades the country's weather-predicting groundhogs against what February and March temperatures actually did, rates Punxsutawney Phil 35% accurate over the 20 years ending in 2024 — good for 17th place out of 19. A bronze statue of a prairie dog in Wyoming outscored him. Over his full span since 1887, independent tallies land near 35–41%, which is to say: worse than flipping a coin.
Everyone repeats that number. Almost nobody shows why it's so low or checks it against the thermometer. We hold the record — 139 million NOAA observations — so we pulled the last five Groundhog Days and scored Phil's call against the observed February–March temperature across the country. He went 2-for-5. And the misses all rhyme: he keeps predicting a long winter that doesn't come.
The Catch: He Almost Always Sees His Shadow
Here is the single fact that explains Phil's whole record. Of 129 predictions since 1887, he has seen his shadow and called for six more weeks of winter 109 times. He has predicted an early spring just 20 times. (One year, 1943, was a wartime no-call, and about ten of the earliest years have no surviving record.) That's a long-winter call in roughly 85% of recorded years.
A prediction that lands the same way 85% of the time isn't a forecast; it's a default. And it's the worst possible default for the climate we actually have. Late winter and early spring across the US have trended warmer over the past half-century — the same warming you can see in our data on the cities where winter is warming fastest. So the one call Phil almost never makes — early spring — is increasingly the one that comes true.
Phil vs. the Thermometer: the Last Five Years
We took each of the last five Groundhog Day calls and scored it the way NOAA does — against the actual February–March temperature. For a 12-city national basket (New York to Seattle, Minneapolis to Miami), we averaged the Feb–March temperature each year from our NOAA records and compared it to the 1991–2020 norm. A long-winter call verifies only if late winter runs colder than normal.
| Year | Phil's call | Feb–Mar vs. normal | Verdict | What happened |
|---|---|---|---|---|
| 2021 | Six more weeks of winter | −0.3°F | Right | Feb–March ran a hair below normal — his one clean long-winter hit. |
| 2022 | Six more weeks of winter | +0.4°F | Wrong | Warmer than normal. No long winter. |
| 2023 | Six more weeks of winter | +1.2°F | Wrong | Warmer than normal again. |
| 2024 | Early spring | +4.1°F | Right | One of his rare no-shadow years — and Feb–March came in far warmer than normal. A clean hit. |
| 2025 | Six more weeks of winter | +2.2°F | Wrong | Warmer than normal across the basket; only Seattle ran cool. |
Two hits in five tries — 40% — which lines up almost exactly with the 35% NOAA finds over 20 years. But look at how he got his two hits. In 2021 the country ran a hair cold, so his reflexive “long winter” happened to land. In 2024 he broke pattern and called an early spring — and Feb–March came in a remarkable 4.1°F above normal, his cleanest hit of the five. Every year he defaulted to winter and late winter ran warm (2022, 2023, 2025), he missed. The groundhog does best on the rare occasions he bets against himself.
This is a 12-city national spot-check against the 1991–2020 average, not an official national index — it's an illustration that happens to match NOAA's independent 35% figure, not a competing verdict. The verdict is NOAA's.
Why No Groundhog Can Forecast Six Weeks Out
Even the best supercomputer models lose useful day-by-day skill somewhere around a week to ten days out — the atmosphere is chaotic, and small errors compound fast. “Six more weeks” is well past that horizon, and a single sunny or cloudy morning on February 2 carries no information about it at all. The shadow test isn't a weak forecast; it's a coin whose edges are weighted toward “winter.”
What does carry real signal six weeks to three months out is the slow-moving state of the oceans — above all El Niño and La Niña — which tilt the odds of a warm or cold, wet or dry season. That's the basis of the only seasonal forecast worth planning around, and it's also why “an official record” is a surprisingly technical thing to pin down; we walk through that in how a weather record is actually verified.
A Fun Tradition — and a Few Better Groundhogs
Phil has been emerging from his burrow on Gobbler's Knob every February 2 since 1887, making him the longest-running weather-forecasting groundhog in the country — which is exactly why NOAA, even while ranking him 17th, wrote that it “would be remiss not to honor” him. The tradition is the point; the forecast is theater.
If you insist on a groundhog, pick a better one. In NOAA's rankings, New York's Staten Island Chuck topped the list at 85% — and the runner-up, at 80%, was Lander Lil, the aforementioned bronze statue of a prairie dog that sits outside a post office in Lander, Wyoming. Georgia's General Beauregard Lee placed third at 75%. The honest reading isn't that Chuck has a gift — it's that with 19 prognosticators making a binary call, somebody lands near the top by chance every few decades, the same way somebody always wins an office bracket.
If You Want an Actual Late-Winter Outlook
Here's the part a groundhog can't give you: this isn't a normal winter. A genuine, measurable Super El Niño is setting up, and decades of past El Niño winters tell you what that tends to do, region by region — which is a real forecast you can check, not a shadow. Start with NOAA's official 2026–27 winter outlook and what the El Niño analogs show, and if snow is your question specifically, see what a Super El Niño actually does to snowfall (short version: the northern snow belt usually gets less, the Mid-Atlantic more). And if you like the idea of grading a forecast against the record, we already did it to a 200-year-old institution: is the Farmers' Almanac accurate? (also a coin flip.)
Caught a Groundhog Day prediction in your town and want to know whether it actually verified? Look up the real February and March temperatures for any US city and year on our weather history lookup — and if your local groundhog beats Phil's 35%, tell us.
Frequently Asked Questions
How accurate is Punxsutawney Phil?
Not accurate as a forecast. NOAA rates Phil 35% accurate over the 20 years ending in 2024, ranking him 17th of 19 US weather-predicting groundhogs, graded against actual February and March temperatures. Across his full history since 1887, independent tallies put him near 35–41%. Our own check of the last five Groundhog Days against the NOAA temperature record has him right just 2 of 5 times — about 40% — worse than a coin flip.
How often does Punxsutawney Phil see his shadow?
Almost always. Of 129 predictions on record since 1887, Phil has seen his shadow and called for six more weeks of winter 109 times and predicted an early spring only 20 times (one year was a wartime no-call, and about 10 early years have no surviving record). That is roughly an 85% long-winter rate — which is the real reason his accuracy is poor. He is not forecasting; he is defaulting to "more winter" nearly every year.
Has Punxsutawney Phil ever predicted an early spring?
Yes, but rarely — only about 20 times in 129 years. His most recent early-spring calls came in 2020 and 2024, and both turned out to be correct: February and March ran warmer than normal both years. His 2025 call reverted to the usual "six more weeks of winter," which did not verify, as late winter again came in above average.
Is Groundhog Day a real weather forecast?
No. There is no physical mechanism by which a groundhog seeing its shadow on February 2 could forecast the next six weeks of weather. Skillful day-by-day forecasting runs out well short of six weeks, and a single February morning carries no information about March. For an actual late-winter outlook, use NOAA’s Climate Prediction Center, which issues probability-based seasonal forecasts and updates them every month.
Sources & Method
Phil's 35% accuracy, the 17th-of-19 ranking, and the other groundhogs' scores are from NOAA's “Grading the Groundhogs”, which scores each groundhog against actual US February and March temperatures over the 20 years ending 2024. The 109-to-20 shadow tally, the 129-prediction record since 1887, and the ~35–41% full-history accuracy range are from the documented prediction log compiled by the Punxsutawney Groundhog Club via Wikipedia. Recent-year calls (2020 and 2024 early spring; 2021–2023 and 2025 six more weeks) are from NBC News reporting. Our year-by-year scorecard is computed from NOAA GHCN-Daily via our database: for a 12-city national airport basket, the mean daily temperature over February–March each year, compared to each station's 1991–2020 Feb–March mean (bad readings excluded). See our data sources & methodology page. Verified October 20, 2026.