Data story · ~130,000 wine reviews

Wine Score Inflation

Wines are rated on a "100-point scale." So why does it feel like everything scores around 90? I ran the numbers on 130,000 professional reviews.

By Jeremy Lee

the hard floor — nothing lower is ever published

86%

of all wines score between 84 and 92

3.1 pts

gap between the most and least generous reviewers

A 100-point scale that's really an 8-point scale

Wine is scored from 50 to 100, but in practice that scale barely exists. Across ~130,000 Wine Enthusiast reviews, nothing is published below 80, the average is 88, and 86% of all wines fall between 84 and 92. Almost the entire "100-point" range is dead space. When failing grades are never printed and everything clusters in an 8-point band, a high score stops meaning much.

Histogram of wine points, floored at 80 and bunched around 88 — Every wine's score. A hard wall at 80 and a tight pile around 88 — grade inflation, in one chart.

Money buys points — with steep diminishing returns

Price and score are related (correlation 0.61 against log-price), but you pay dearly for each point. A roughly $10 wine averages about 85 points; a $100 wine averages about 92. Spending ten times as much buys you, on average, about seven points — and most of those points come early.

Average points rising with price on a log scale, then flattening — Average points by price (log scale). The curve rises fast then flattens — diminishing returns on every extra dollar.

It's not a "recent vintages are better" effect

You might guess the scores have crept up over time. They haven't: average points are essentially flat across vintages (about 0.09 points per decade). The compression isn't a recency trend — it's baked into how the scale is used.

Average points by vintage, nearly a flat line — Average score by vintage year — basically a flat line. The inflation is structural, not generational.

The reviewer matters as much as the wine

Who tastes the wine moves the number. Among the most prolific reviewers, average scores range across 3.1 points — roughly the same boost you'd get by paying ten times more for the bottle.

Bar chart of average points by reviewer, spanning about 3 points — Average points awarded by the ten most prolific reviewers — a real, systematic spread.

The findings

Scores are floored at 80 and 86% land in 84–92: the 100-point scale is effectively ~8 points wide.
Points track log-price (r = 0.61) with steep diminishing returns: ~$10 → ≈85 pts, ~$100 → ≈92 pts.
Average score is flat across vintages (~0.09 pts/decade) — the compression is structural.
The reviewer adds a 3.1-point spread — as much as a 10× price jump.

What to do with a wine score

Treat the number as a coarse signal, not a verdict. The gap between an 88 and a 90 is mostly noise — vintage, reviewer, and rounding. If you're hunting value, the early part of the price curve is where the points are cheapest, which is exactly the thread I pull on in my New World vs Old World wine value work.

Data & method: the public "winemag-data-130k-v2" Wine Enthusiast dataset (~130k reviews; Kaggle, via a GitHub mirror). Vintage parsed from each wine's title. Caveat: this is a single snapshot from one publication, so it can't measure score drift by review date over time — the "inflation" here is the structural compression of the scale (an 80 floor and a narrow band), plus the price, vintage, and reviewer effects, not a year-over-year creep. Reproduce with fetch_data.py + analyze.py in the repo.