Across 7,045 regular-season NFL games from 1999-2026, the single most common final margin is three points. It happens in 15.00% of games — 1,057 of them — which is 2.80 times what a smooth bell curve of the same width would predict.
That is not a small distortion. It is the central fact of how NFL games finish, and it has a price.
The distribution
Margins are not spread evenly, and they are not spread normally. Here is every margin from a tie to 17 points, against what a normal distribution with the same standard deviation (14.60 points) would put there.
| Margin | Games | Share | Smooth curve | Ratio |
|---|---|---|---|---|
| 0 | 16 | 0.23% | 2.73% | 0.08x |
| 1 | 294 | 4.17% | 5.45% | 0.77x |
| 2 | 291 | 4.13% | 5.41% | 0.76x |
| 3 | 1,057 | 15.00% | 5.35% | 2.80x |
| 4 | 346 | 4.91% | 5.26% | 0.93x |
| 5 | 253 | 3.59% | 5.15% | 0.70x |
| 6 | 421 | 5.98% | 5.02% | 1.19x |
| 7 | 642 | 9.11% | 4.87% | 1.87x |
| 8 | 268 | 3.80% | 4.70% | 0.81x |
| 9 | 113 | 1.60% | 4.52% | 0.35x |
| 10 | 393 | 5.58% | 4.32% | 1.29x |
| 11 | 161 | 2.29% | 4.11% | 0.56x |
| 12 | 117 | 1.66% | 3.90% | 0.43x |
| 13 | 192 | 2.73% | 3.68% | 0.74x |
| 14 | 337 | 4.78% | 3.45% | 1.39x |
| 15 | 106 | 1.50% | 3.22% | 0.47x |
| 16 | 148 | 2.10% | 3.00% | 0.70x |
| 17 | 247 | 3.51% | 2.77% | 1.26x |
These 18 margins account for 76.7% of all games. The rest is the blowout tail, where individual margins are rare enough to be uninteresting on their own.
Three things in that table are worth sitting with.
Three is 2.80 times its fair share. A field goal is the most common scoring play and the most common margin of victory, and the two facts are the same fact.
Seven is the second spike, at 9.11% against a smooth 4.87%. A touchdown and the conversion that usually follows it.
Ties are nearly extinct. A smooth curve says 2.73% of games should finish level. The real figure is 0.23% — 16 games. Overtime resolves almost all of them, which makes zero the one margin the bell curve overstates by more than tenfold.
And the numbers between the spikes are suppressed. One point happens 4.17% of the time and five points 3.59%, both below their smooth share, because the ways a football game ends do not distribute themselves evenly.
What a half-point is worth
This is where the distribution stops being trivia. If you move a spread across a key number, you hand the other side every game that lands exactly on it.
Take an underdog from +2.5 to +3 and you gain the 15.00% of games decided by exactly three. On a distribution that is otherwise symmetric, that turns a coin flip into 57.5% — a fair price of -135.
| Number | Games landing there | Dog wins at the whole number | Fair price (dog) | Fair price (favourite) |
|---|---|---|---|---|
| 3 | 15.00% | 57.5% | -135 | +135 |
| 7 | 9.11% | 54.6% | -120 | +120 |
| 6 | 5.98% | 53.0% | -113 | +113 |
| 10 | 5.58% | 52.8% | -112 | +112 |
| 14 | 4.78% | 52.4% | -110 | +110 |
Read the first row as the one that matters. Buying half a point onto 3 is worth about -135 in fair odds; selling it is worth +135. At 7 the same trade is worth -120. Below that, at 6 through 14, it is worth having and not worth much.
The teaser everybody sells you
The best-known application of all this is the six-point teaser that moves an underdog from +1.5 to +7.5, crossing both 3 and 7. It is recommended constantly, usually as a standing edge.
It crosses 42.7% of the distribution, which takes a leg from a coin flip to about 71.4%.
A two-team teaser priced at -110 needs 72.4% per leg to break even. At -120 it needs 73.9%.
So it comes up 1.0 points short at the friendlier of those two prices, and further short at the other. The move is real, the numbers behind it are real, and it still does not clear the toll. That was likely not always true — teasers were cheaper once — but it is true of the distribution as it stands and the prices as they are posted.
What this does not mean
It would be easy to end here by telling you to go and buy half-points, and that would be wrong.
We also measured what books actually charge for alternate spreads: across 7,045 games of distribution and a full slate of posted alternate prices, the margin built into those prices is roughly flat — about three percent a side — at every distance from the main number, and no thinner at 3 or 7 than anywhere else. The clustering is real and the books know about it. They have known for decades.
What the table is good for is understanding what you are buying. A half-point is not a rounding error, it is 15.00% of all outcomes at one specific number and under two percent at most others. If you are going to pay for it, pay for it where it is worth something, and know roughly what it should cost.
Method
Every figure above comes from 7,045 completed regular-season games, 1999-2026, each one carrying a closing spread and a final score. Margins are absolute, so a three-point win and a three-point loss are the same row. The smooth-curve column is a normal distribution centred on zero with the measured standard deviation of 14.60 points, evaluated over the one-point band around each margin and doubled for the two signs — except at zero, which has one.
The mean absolute margin across the sample is 11.53 points.
Measured 2026-10-13. This piece is generated from the data rather than written around it, so it can be regenerated as the sample grows, and the figures will move slightly when it does.
