The Buffalo Bills' 2025 defense is two defenses stitched together. One of them was excellent — a top-ten pass defense with the single best behind-the-line coverage unit in the NFL. The other was one of the four worst run defenses in football, and by season's end opposing coordinators had figured that out and were openly game-planning around it.

Table of Contents

1. The two-defense problem, in one line
2. The pass defense that actually got better
3. The behind-the-LOS strangler
4. The run defense that fell off a cliff
5. Short yardage: the specific failure mode
6. Edge runs, and the disappearing stuff
7. The tell: opponents are already re-scripting
8. What I'd watch next

1. The two-defense problem, in one line

Buffalo's team-season EPA splits from 2023 through 2025 tell you almost everything up front:

The pass number swung by roughly 15 hundredths of an EPA point in the good direction. The rush number swung by roughly 15 hundredths in the bad direction. Buffalo essentially traded run defense for pass defense — and 2025's +0.085 rush EPA/play allowed was the fourth-worst mark in the NFL, trailing only the Giants (+0.167), Cowboys (+0.095) and Bengals (+0.087).

2. The pass defense that actually got better

At −0.091 pass EPA/play allowed in 2025, Buffalo ranked ninth-best in the league — sandwiched between Philadelphia (−0.105) and the Chargers (−0.089). This is the same unit that was +0.062 (a bottom-third pass defense) in 2024. That's a meaningful improvement, and it happened despite the pass rush being basically the same box score:

A tick up on both, but not the kind of jump that would explain 15 hundredths of EPA/play. The improvement came in coverage — and, specifically, in one very particular zone of the field.

3. The behind-the-LOS strangler

Any pass thrown at or behind the line of scrimmage in 2025 against Buffalo was a disaster for the offense. On 106 such attempts, opponents completed just 66.98% and threw 4 interceptions, and the plays averaged −0.524 EPA/play.

That's not "good." That's league-leading. Across all 32 defenses in 2025, Buffalo ranked 1st in behind-the-LOS EPA allowed, narrowly ahead of Indianapolis (−0.505) and comfortably ahead of Minnesota (−0.398), Houston (−0.379), and Detroit (−0.378).

Compare that to 2024, when the same defense saw 151 behind-LOS targets, allowed a 75.5% completion rate, and gave up −0.062 EPA/play — decent, not remarkable. The volume dropped and the punishment quadrupled. If your checkdown against Buffalo was a screen, a swing, or a dumpoff, you were basically taking a negative-half-a-point on the play before you snapped it.

Four interceptions on 106 low-air-yards attempts is also a heavy interception density in a bucket where picks are usually rare — these are throws that get tipped, jumped, and turned into turnover-worthy plays.

4. The run defense that fell off a cliff

Now the other half. Buffalo's rush defense in 2024 was already merely average by EPA (−0.065/play), but the underlying numbers were fine — 22.8% of runs were stuffed at or behind the line, opponents averaged 4.58 yards per carry, and Buffalo bled a totally reasonable amount of ground production.

2025 broke all of it:

Every direction the numbers moved in was bad. Yards per carry up more than half a yard. Stuff rate down six full percentage points. EPA/play up almost 19 hundredths.

5. Short yardage: the specific failure mode

The nastiest bucket is short yardage — runs with 3 or fewer yards to go for the first down or touchdown. In 2024, opposing offenses tried this 79 times against Buffalo and converted 51.9% of the time, at −0.146 EPA/play. That's a good short-yardage defense.

In 2025, on effectively the same volume (80 attempts), opponents converted 76.3%, at +0.419 EPA/play.

Zoom in even tighter, to the sneak/dive zone at 1 yard to go: the conversion rate allowed jumped from 60.9% (28-of-46) in 2024 to 85.0% (34-of-40) in 2025, and the EPA per attempt swung from +0.013 to +0.547. Buffalo's short-yardage defense didn't slip — it inverted. Teams stopped fearing 3rd-and-1 against the Bills, and they were right not to.

6. Edge runs, and the disappearing stuff

The failure has a directional signature too. Splitting run attempts by tracked run location:

In 2024 Buffalo still had a solid middle (−0.126 EPA/play, 3.85 YPC) that anchored the defense even as the edges leaked. In 2025 the middle collapsed to +0.076 EPA/play and 4.84 YPC — from a strength to a below-average number — and the edges got worse on top of it. Combined edge (left + right) went from 302 attempts at 4.97 YPC / −0.012 EPA in 2024 to 337 attempts at 5.39 YPC / +0.118 EPA in 2025.

You can't have both the interior giving up nearly five yards a pop and the perimeter giving up nearly five and a half. That's what "the run defense broke" looks like at the play level.

7. The tell: opponents are already re-scripting

If the run defense were quietly bad but nobody noticed, that would be one thing. But the play-calling against Buffalo has shifted, hard:

Buffalo faced the run on roughly 47% of pass/run scrimmage plays in 2025 — nearly 7 percentage points higher than two years earlier, and more than 5 points higher than a year earlier. On 950 pass/run plays that's an extra ~50 rushes redirected toward the Bills relative to their 2024 diet, and closer to ~65 relative to 2023.

That's exactly the strategic reaction you'd expect from opposing staffs watching the same film. Buffalo will make you regret every checkdown you throw — behind-the-LOS EPA of −0.524 says so — but they'll also let you run into a first down on 3rd-and-2. So teams stopped throwing checkdowns and started running. The pass rush numbers barely budged (sack rate 6.3% → 7.1%, QB hits 14.9% → 17.0%), which means the reduced dropback volume against Buffalo isn't coming from more pressure. It's coming from opponents choosing not to drop back.

8. What I'd watch next

A couple of caveats worth stating in plain language. Rush EPA is noisy at the season level, and one blown gap on a 60-yard run can move a team's number a lot. But the mechanism here is not one big run — it's short-yardage conversion rate jumping 24 points, stuff rate falling 6 points, and edge runs adding half a yard per carry across 337 attempts. That's a structural issue, not a variance blip.

Two things I'd be watching if I were a Buffalo fan or a Week 1 opponent in 2026:

• Does short yardage stay broken? A 76% conversion rate allowed on 3-and-shorter is the kind of number that either signals real personnel/scheme rot or regresses hard. It's the fastest single lever available to fix the defense's EPA profile.
• Do offenses keep dialing up the run share? A 46.84% run rate against a defense whose pass unit is top-ten in EPA is not an accident. If that stays elevated into 2026, it's the market telling you the pass-defense-strong / run-defense-weak split is real and priced in.

Every "great pass D, one weakness" story from 2025 has been about a coverage-shell mismatch — the deep ball against Minnesota, over-the-middle windows against Pittsburgh, whatever. Buffalo's is different. Their weakness isn't in the passing game at all. It's the run game itself, at the point of attack, on the downs where offenses have the most control over what they call. And unlike most of these stories, this one has already stopped being a secret.