🧪 Is EuroMillions really random?
We run all 1,990 EuroMillions draws since 13.02.2004 through 11 statistical tests. The question: do the draws behave the way pure chance predicts, or are there patterns, “hot” numbers, seasonal effects?
No anomaly: 0 of 11 tests fall below the 5% threshold – pure chance would produce 0.55. After correcting for multiple testing (Holm), no finding remains.
| Test | Finding | p-value | p (Holm) |
|---|---|---|---|
| Uniform numbers Is every number drawn equally often? χ² = 58.1 with 49 df; largest deviation: 22 (z = -3.06). | normal | 0.175 | 1.000 |
| Uniform Lucky Stars Are all Lucky Stars drawn equally often? 1,050 draws; χ² = 15.2 with 11 df; largest deviation: 2 (z = 2.40). | normal | 0.175 | 1.000 |
| Seasonal numbers Do some numbers come up more often in certain months? number × month, χ² = 613.7 with 588 df; most extreme cell: 42 in August (z = 4.05). | normal | 0.224 | 1.000 |
| Draw day Do the numbers depend on the day of the draw? number × weekday, χ² = 92.4 with 98 df; most extreme cell: 22 on Tuesday (z = -2.65). | normal | 0.641 | 1.000 |
| Stability over time Have the frequencies shifted over the years? 4 periods, χ² = 217.9 with 196 df; largest deviation: 50 in period 1 (z = 3.04). | normal | 0.135 | 1.000 |
| Sum of numbers Do the sums follow the exact distribution? mean sum 127.53 (theoretical 127.50); χ² = 41.6 with 31 df. | normal | 0.096 | 1.000 |
| Even/odd Is the mix of even and odd numbers right? χ² = 5.9 with 5 df. | normal | 0.313 | 1.000 |
| Consecutive numbers Do pairs like 12–13 appear as often as expected? 34.8% of draws contain consecutive numbers (theoretical 35.3%). | normal | 0.621 | 1.000 |
| Repeats from previous draw Does a draw depend on the previous one? on average 0.49 numbers of the previous draw come up again (theoretical 0.50). | normal | 0.685 | 1.000 |
| Number pairs Do certain pairs come up together too often? most frequent pair 4 + 23: 29 times (expected 16.2); spread of all pairs compared with 100 random series. | normal | 0.376 | 1.000 |
| Gaps between hits Are a number's absences as long as pure chance predicts? gap distribution compared with 100 simulated random series. | normal | 0.891 | 1.000 |
The latest draw
The draw of 16.10.2026 (7 15 16 22 35) has no unusual feature.
Almost every draw has some “rare” feature if you look at enough features. That is no sign of manipulation.
Strategies on trial: do frequent or overdue numbers predict anything?
For every draw since 13.02.2004, the test picks numbers using only the earlier draws (last N draws) and counts how many come up in the next draw. The benchmark is what a random selection of the same size hits on average.
Popular strategies
| Strategy | Selection | Hits per draw | expected | ratio | p-value | p (Holm) | |
|---|---|---|---|---|---|---|---|
| … | |||||||
Test your own mix
N1/N2: frequency within the last N draws. N3/N4: current gap (draws since last hit, over the whole history). Overlaps count once.
Distribution of hits per draw
observed expected (chance)
Cumulative excess / shortfall vs. chance
Important: every combination has the same chance in every draw. No strategy can improve the number of hits; test enough variants and you will still find some with p < 5% – that is chance (see above, multiple testing).
How to read the table
The p-value is the probability of a result at least this extreme under pure chance. Below 5%, a single test is usually called “significant”. Because 11 tests run at once, about 0.55 such hits are expected by chance alone; the Holm correction accounts for that. Only a result still below 5% after correction would be a serious hint.
Method
Based on the official draws of the current 5/50 format (since 13.02.2004). Frequency tests: chi-square corrected for drawing without replacement. Pairs and gaps: compared with 100 simulated random series of the same length (Monte Carlo, two-sided). Everything is recalculated after every draw.
Rules & FAQ
Is EuroMillions rigged?
The data show no sign of it. 0 of 11 tests fall below 5% (0.55 expected by pure chance), 0 after correcting for multiple testing. The draws behave like those of a fair drawing machine.
Can these tests predict winning numbers?
No. Precisely because the draws pass the tests, there is no exploitable pattern: every combination has the same chance in every draw.
Why is a test sometimes “unusual”?
At the 5% level, one test in twenty fires by pure chance on average. That is why we also show the Holm-corrected p-value, which accounts for the number of tests.