The best quarterbacks of all time

And our projected best (and worst) QBs of 2025. Introducing QBERT ratings and previewing our ELWAY NFL model.

If viewing this in the Substack app or on email, we strongly suggest that you check out the web version of this article instead since it has a whole bunch of fancy charts.

For roughly the past six weeks, I’ve been immersed in developing an NFL projection system for Silver Bulletin subscribers, which we’re calling ELWAY¹. We’ve analyzed every professional game since literally the start of the American Professional Football Association in 1920. I’ve written a couple thousand lines of code. I’m really excited about it: the FiveThirtyEight NFL models were good, but ELWAY will go a lot further in various dimensions. It’s about 85 percent ready.

But can I be honest? The NFL season starts on Thursday, and there’s obviously just not enough time left to get it published by then no matter how many more 4 AM coding sessions I put in. Plus, for any “rookie” model, it’s always nice to take some time to see how it behaves when presented with fresh, out-of-sample data. So, we’re going to roll out ELWAY in stages:

• The quarterback component of ELWAY — QBERT² — is pretty cool on its own; really a whole model unto itself. And QBERT is done, so that’s what you’re seeing first. This article introduces QBERT with a lot of historical perspective on quarterbacks.

• As a bonus for paying subscribers, this article also includes our initial 2025 quarterback projections — QBERT predicts future ratings in addition to assessing past performance — plus a spreadsheet of historical QBERT ratings.

• We’ll also create a landing page for QBERT after a week or two, which will be updated weekly.

• The rest of ELWAY — including projected W-L records and playoff odds for each team, plus game-by-game projections for scoring margins, win probabilities and total points scored³ — is coming soon: we’re hoping by two to four weeks into the season. ELWAY projections incorporate QBERT and other factors including roster turnover and detailed data on each game that goes far beyond the final score — much of which are more predictive of future performance than plain ol’ points scored and allowed. Most features of ELWAY will also be for paying subs.

We’re in this for the long term, and we want everything that we do at Silver Bulletin be among the best versions you can find anywhere — even if that sometimes means they take a little longer to get ready. Now, here’s QBERT.

---

Why the world needs a better QB rating system

Not to get all anthropological on you, but as the most prominent position in the country's favorite sport, NFL quarterbacks occupy a role like almost nothing else in the American psyche. There are 32 starting quarterbacks in the NFL, as compared to 50 governors and 100 senators — but the average American could probably identify more signal-callers than log-rollers. It’s the quarterbacks who are the household names, who carry a city's hopes and dreams on their shoulders in a culture that has fewer and fewer focal points left.

And there's a good reason for that. There's a line of thinking that a QB is just one player on an 11-man offense and 53-man roster, that he gets unearned credit or undeserved blame for whatever his team does. In researching ELWAY, however, we've discovered that the contrarian take is wrong. Quarterbacks, as Vegas oddsmakers long ago figured out, are really damned important. We've been able to “discover” this in various ways, but mostly by observing what happens over the past 75 NFL seasons — our QBERT ratings date back to 1950 — when a QB is injured or joins a new team. We estimate that quarterbacks are responsible for something like one-third of all player value generated in today’s NFL, far more than the share of the money they receive under the salary cap (about 12 percent, subtracting out minimum salaries).

While, for the most part, QBERT is aligned with conventional wisdom — not necessarily a bad thing — we believe it gives credit for elements of QB performance that other systems overlook. The NFL’s traditional passer rating assigns no credit at all for QB rushing plays, even though QBs from Steve Young to Lamar Jackson have completely disrupted opposing defenses that way: ignoring this isn’t much better than neglecting the impact of 3-point shooting in today’s NBA. Passer rating also fails to incorporate basic statistics like sacks, fumbles, and first downs generated (even though first downs are in some ways the backbone of the sport). And it makes no attempt to allocate credit between a QB and his receivers and offensive line — and here’s a more subtle one: QBs are also more efficient when they have better running backs — while QBERT does.

ESPN’s QBR is considerably better — in fact, QBERT ratings are bootstrapped to QBR in various ways. But it nevertheless has some limitations. It probably gets too cute about trying to give credit for performance in high-stakes, high-leverage situations — which, as important as they are, can inherently be random and introduce noise into their ratings. QBERT is more consistent from start to start, and actually predicts QBR as well as QBR itself.⁴

Passing efficiency has improved dramatically, inflating the ratings for modern QBs

And QBR is only available dating back to 2006, whereas we’ve calculated QBERT ratings for every game since 1950. Once you do that, you’ll realize that you need a system where the greats of the past can compete on a level playing field with today’s stars. Nearly everything that has happened in the NFL — from bigger offensive lines to better sports medicine to officiating rules that increasingly protect the quarterback — has favored perpetual growth in passing efficiency. Other than a few greats like Young, Joe Montana, and Dan Marino, MVP-type seasons from most quarterbacks before 2000 would grade out as merely good today.

However, our adjusted QBERT ratings account for this by calculating a rolling baseline of QB performance that every player’s raw performance is continuously benchmarked against. We also adjust for the quality of opposing defenses faced, home field advantage, and even the weather: cold-weather QBs like Brett Favre faced a significant disadvantage as compared to those who spent their career in domes or in balmy climes.

If you noticed that unadjusted QBERT ratings closely track with traditional passer rating, that’s intentional. One thing we like about NFL passer rating — at least as it existed in the 1980s and 1990s before further inflation in passing statistics made triple-digit ratings common — is that it follows a fairly intuitive scale. A QB rating in the 90s was great, something in the 80s was good, whereas anything in the mid-70s or below meant the team was left wondering if this guy taking every offensive snap was really the franchise QB they hoped for when they drafted him. Adjusted QBERT ratings are designed to follow the same benchmarks:

We were also able to derive replacement level in a rather literal way: it’s defined by what happens when an undrafted rookie makes his first NFL start. Although some of those no-name undrafted starters might later turn out to be Tony Romo or Warren Moon, most of the time they’re pretty bad. Ironically, however, the first start for highly touted rookies — even #1 overall picks — doesn’t necessarily go any better. Quarterbacks face a steep learning curve, and high picks like Ryan Leaf are often given opportunities before they’re ready for prime time, or even after they’ve proven they aren’t.

Because it’s such a difficult position, the league can’t help but to take a sink-or-swim approach with its quarterbacks. Zach Wilson (like Leaf, the #2 overall pick) turned out to be a goat, while Tom Brady (#199) turned into the GOAT. (Yes, Brady is tops in our system.) It’s rare for quarterbacks to hold their starting jobs with ratings below the mid-70s, though high draft picks will get more leash than others.

How QBERT ratings work

Having given you a few teasers about where quarterbacks rank, let me briefly explain more about how QBERT ratings are calculated. If you don’t care about this and want to see the top seasons and careers — well, just scroll down. Or if you really care, there are a lot of footnotes, and we’re working on a methodology page to come soon.

While I like ESPN QBR, it’s something of a black box. However, through regression analysis, we were able to essentially crack the code and explain about 75 percent of the variation in game-by-game QBR based on statistics that are more widely available, in most cases stretching back for decades.⁵ (We’re also not convinced there’s necessarily much signal in the remaining 25 percent.) These are the basis for QBERT, and here are the factors it considers⁶:

Primary factors

• Passing and rushing TDs

• Completions

• Interceptions

• Fumbles⁷

• Passing yards before the catch (a.k.a. Air Yards) — our analysis suggests that credit for yards after the catch should mostly go to receivers⁸

• Net QB rushing yards (rushing yards gained less sack yardage lost). This is a big factor that’s neglected by other metrics that focus on passing efficiency alone. It accounts for much of the difference between QBERT and the conventional wisdom. Both rushing yardage and sacks are really important, especially when taken together.⁹

• Passing and rushing first downs¹⁰

Secondary factors

• QB rushing attempts (fewer is better, holding other factors constant¹¹)

• Pass pressure: a quarterback gets more QBERT credit for facing a lot of it¹²

• Running back yards per carry: a QB gets less credit if his running game is effective¹³

• Game script, as measured by the net score at end of the second and third quarters¹⁴

• And clutch wins, defined as any win in overtime, or when the QB’s team trailed entering the fourth quarter. This has a fairly minor impact; QBERT isn’t too “clutch-pilled”.¹⁵

This gives us a raw QBERT rating, which is then tweaked for various circumstances to create adjusted QBERT:

• The strength of the opposing defense; each defense basically gets a rolling Elo rating based on the QBERT it has allowed in recent games

• Home field advantage

• Weather: QB performance declines substantially in cold temperatures¹⁶ or especially with wind

• And most importantly, a rolling factor that tracks the change in leaguewide QBERT scores over time, so the average always hugs the 80 line closely.¹⁷

Just how valuable are the best quarterbacks?

Last thing before we finally address the question implied by the headline. Since QBERT is an efficiency measure — how good the QB is per snap? — a quarterback’s overall value in a game, season or career is determined both by his adjusted QBERT rating relative to the replacement level of 68, and by his number of QB plays, defined as passing attempts + sacks + QB rushing attempts excluding kneels. Since 1950, when QBERT begins, there have been 15,787 total NFL and AFL games, counting the playoffs. A team full of replacement-level players would be very bad: we estimate about 1-15 over a 16-game schedule.¹⁸ After subtracting these out, we allocate 30 percent of the remaining wins to quarterbacks in the form of wins above replacement (WAR). As the rushing game has become less prominent, in fact, this percentage has gradually been increasing to around 33 percent today.

How did we get to that 30 percent? It’s sort of a consensus figure. We ran some complicated statistical analyses on what happens when QBERT ratings are added to the team-based measures from ELWAY. These suggest that, if anything, 30 percent is too low and the value might be closer to 40 percent instead. We looked at Vegas point spreads, because QB changes have a similar effect in ELWAY as they do on Vegas lines, and tried to back out some estimates from there. And we checked whether this passed a “smell test”. We estimate that the best QB seasons in NFL history are worth something like 6 to 8 WAR in the regular season¹⁹ — so if you replaced Jackson or Mahomes or Josh Allen with a scrub, their teams wouldn’t be terrible but would probably be an underdog to finish over .500.

I’d be more comfortable if you asked me for a range: I’d say that QBs are probably responsible for somewhere in the range of 25 to 40 percent of all marginal value generated by NFL players. If you think I’m wrong, then Vegas disagrees with you, too, and you can make some very profitable bets whenever a star QB gets hurt.

One objection you might raise is that this credits QBs with a huge share of whatever happens on offense. If offense is 45 percent of football, defense is 45 percent and special teams are 10 percent, there’s almost no room left to give credit to receivers, offensive linemen and running backs.

Except, offense is worth more than that. Our analysis for ELWAY suggests there’s considerably more predictability in points scored per game than points allowed. Moreover, offense bleeds into defense — and vice versa — because holding the ball denies the opposing team the opportunity to score and sticks them with worse field position. In ELWAY, we attribute around 62 percent of overall marginal value to offensive players, 36 percent to defensive players and just 2 percent to special teams.²⁰

The top 100 quarterbacks by QBERT

Since I just wrote at length about WAR, let’s start there. To head off this question, the reason why Mahomes isn’t rated higher is simple: he’s in the middle of his career and he’s on track to be among the best ever once he’s finished. Still, modern QBs do have some advantages in WAR: there are more regular season games now, more playoff games, and quarterbacks account for a higher share of offense than they once did.

In career value, Brady is #1 with a bullet, and it’s not particularly close; he has 114.3 lifetime WAR versus 94.8 for Peyton Manning and 86.5 for Drew Brees. And sure, that’s partly due to his longevity — on an efficiency basis, others like Marino and even Dan Fouts are similar, not to mention active QBs like Mahomes and Jackson. But when you win one Super Bowl in 2002 and another in 2021 (and five more in between), you’re the GOAT.

But otherwise, you can see big effects from the era adjustment. You can also detect how much running or at least scrambling ability matters; guys like Steve McNair and Randall Cunningham rank higher here than they probably do in the popular imagination.²¹ The weather adjustment also makes a bigger difference than you’d think.²² As evaluated by QBERT, Fran Tarkenton, who played in Minnesota before the Metrodome was built, rates higher than Philip Rivers, while the reverse is true in Pro-Football-Reference’s approximate value.

As for Mahomes, if you figure he’s halfway through his career and will double his WAR, he’ll eventually rank #3 all time between Manning and Brees — but Brady will be hard for anyone to catch.²³

Here’s the list sorted by QBERT instead, meaning efficiency per play rather than overall value. I’ve set a minimum of 2,000 quarterback plays for this list, although that’s actually not so easy to achieve. NFL quarterbacks have long careers, and only about 150 QBs since 1950 have started at least ten games in at least five distinct NFL or AFL seasons, so you’ll get some distinctly forgettable quarterbacks in the back half of the top 100.

There’s a lot we might say here. Again, the era adjustments make a really big difference. Quarterbacks like Otto Graham, Johnny Unitas and Norm Van Brocklin were actually below average in raw QBERT, meaning basically that a quarterback with those statistics today might struggle to hold down their job. But they were much better than other quarterbacks of their day. Roger Staubach is a more subtle example: he played in a very defense-forward era in the 1970s and he was a good scrambler, so he benefits from a couple of the factors that QBERT considers more prominently than other methods.²⁴

Still, the two biggest outliers are the former 49ers teammates, Young and Montana, who posted stat lines that would fit in perfectly well in the modern game, but decades ahead of time. In unadjusted QBERT, the highest all-time ratings are for Mahomes, Allen and Jackson in that order — but Young ranks fourth in unadjusted QBERT despite predating them by almost three decades.

Next, the top QB seasons by QBERT WAR. These include WAR accumulated in the playoffs; we’ve always considered it a weird blind spot in the NFL and other sports that winning titles is seen as the sine qua non of success, and yet statistics accumulated during post-season games — Brady threw for 13,400 yards in the playoffs! — are all but ignored in historical assessments.

There’s strong correspondence here between what league MVP voters thought at the time and our QBERT assessments. While QBERT doesn’t agree with every MVP pick — it would have chosen Jackson over Allen last year and Brees over Rodgers in 2011 — it also picks up on some forgotten-about seasons like Ken Anderson’s 1981, Matt Ryan’s 2016 and even Brian Sipe’s 1980 that were rewarded with hardware, even if they aren’t fully appreciated today. (Anderson also ranks #17 on the career QBERT WAR list, but isn’t in Canton.)

Once again, though, it’s Brady who tops the list for his 2007 season — yes, that’s the year that the Patriots went 16-0. And, yes, it’s one of the times that Brady lost the Super Bowl: he posted a mediocre 79.8 QBERT against the Giants, in fact. Jackson’s 2024 season ranks 10th all time, although with a 41-to-4 TD-to-INT ratio plus 915 rushing yards, you don’t need an advanced stat like QBERT to see the merits of it even if MVP voters didn’t.

That WAR list is slightly biased toward recent QBs, though, given that there are more regular season and playoff games nowadays. So here are the best seasons ranked by adjusted QBERT, with a minimum of 250 QB plays.

Montana’s 1989, Young’s 1994 and Manning’s 2004 are essentially tied for first as the most efficient seasons of all time, and nobody is going to complain about those. But I don’t like this list as much, with some QBs sneaking onto it with smallish sample sizes: you’ve probably never heard of Babe Parilli before, who had a pretty good year but played just 10 games for the Boston Patriots in 1962. Still, there’s once again strong correspondence with MVP voting.

Tracking and projecting quarterbacks over time

Everything we’ve discussed so far is purely retrospective. But QBERT is also designed to plug into ELWAY, the projection system that we’ll unveil in a few weeks. While I don’t want to spoil everything here — we’re hoping that some of you will sign up for a paid subscription! — let me tease some of the cool things that having evaluated more than 30,000 quarterback starts since 1950 will allow us to do.

Let’s take a great but polarizing quarterback as an example: Rodgers. Here is his forecasted QBERT rating — his prior — for each of his starts since 2008, when he finally supplanted Favre on the Packers. We’re still working out the kinks in how we might visualize this data, but this should get the basic idea across:

Rodgers begins with an expected QBERT of just 69.1 in his first start. But that’s nothing against him because pretty much every quarterback starts out with a mediocre rating. These initial ratings are based on a player’s college statistics — yes, college QBERTs are moderately predictive of NFL QBERTs, provided you adjust them for strength of schedule. And Rodgers’s stats at Cal were good but not great. But great numbers would only help a little bit. Barring some weird cases where a QB took a significant number of snaps as a backup before starting a game²⁵, the players with the best college QBERT ratings (the system really likes Jayden Daniels’ NCAA numbers, for instance) still only project to a rating of around 70 in their first start in the league.

However, QBERT also expects very rapid improvement over roughly a QBs first 20 starts — enough that rookie QBs are typically much better by the end of the year than at the start — provided that the player continues to hold down the starting job.

That’s a vital qualifier, because it’s extremely costly to throw away a season with a sub-average quarterback. So the mere fact that a QB continues to start provides a clue to his trajectory; there is a considerable amount of survivorship bias that you have to navigate around when projecting quarterbacks.

The main exception to this are high draft picks like Leaf and Wilson, for whom two or three seasons might be sacrificed to rationalize the draft pick. For this reason, in fact, draft position isn’t used in QBERT. Empirically, undrafted rookies like Romo start out with almost the same QBERT as first-round picks like Rodgers because in order for them to be granted a start, they have to have shown some promise in practice.²⁶ If every draft pick were guaranteed a fixed number of starts, there would be a stronger correlation with draft slot, but they aren’t.²⁷

After a few seasons, the prior based on a college stats rolls out of the system completely, and a player has to exhibit the improvement that QBERT expects out of a young quarterback. Rodgers did that, posting an adjusted QBERT of 88.2 in his first full season and then continuing to improve until logging a gaudy 108.6 rating in his first MVP season in 2011. While actual QBERT can be noisy, expected QBERT tends to be smoother as the system bakes in an aging curve that also accounts for how many games a quarterback has started recently.

By mid-career, QBERT is more agnostic about a player’s expected trajectory, and you can see Rodgers’ numbers jumping around with peaks corresponding nicely with his other MVP seasons (in 2014, 2020 and 2021). Often, gaining more experience is enough to offset expected aging into a quarterback’s mid-thirties — it’s such a difficult position that you can never really have enough reps — but this isn’t always the case. Rodgers’s late-career renaissance — he won his last MVP at age 38 — was a little bit of a surprise to QBERT. QBs generally age well, but that’s later than you would ordinarily expect.

There will be more about this process on the methodology page. It’s a bit involved. QBs are regressed somewhat toward a long-term mean at the start of a fresh season: sometimes the opportunity to look at film and design a new scheme or tinker with the personnel helps. So while Baker Mayfield arguably had a better season than Mahomes last year (as discussed in SBSQ #23), Mahomes will start out with a considerably higher QBERT projection in Week 1 for 2025. The system is also punitive to players who miss starts, especially if they’re older. You can see that Rodgers was expected to decline from above-average to merely average after missing nearly all of the 2023 season — and that’s pretty much what the Jets got out of him last season. So QBERT isn’t expecting any miracles out of Rodgers’s new team, the Steelers.

Initial QBERT projections for 2025

Here’s how the system ranks every quarterback currently on an NFL roster if they were to start in Week 1.²⁸ I’ve also listed their QBERT ratings over the past three seasons for reference.