How to read a wine score: 100-point, 20-point, star, and percentile systems

Q: Is a 95 Parker score really better than a 92?

Yes, but the gap is smaller than it looks because the effective scale compresses 15 points into 15 ranks. A 95 vs 92 is roughly the difference between top 10% and top 30% of wines rated by Parker — meaningful but not as dramatic as "95 vs 92" sounds in absolute terms.

Q: Why does Rankquant rescale crowd ratings so aggressively?

Because the math requires sources to be on a common scale before weighting. Vivino × 20 brings the crowd 1-5 onto a 20-100 effective range that the source-weighted mean can aggregate with 100-point professional scores. The rescaling is purely mechanical; it's not a claim that a Vivino 4.3 equals a Parker 86.

Q: What about the 5-star "half-star" increments?

Handled as decimals (4.5 → rescaled to 90, 4.3 → 86, etc.) before aggregation. No special treatment.

Q: Do older Parker scores still count the same as recent Vinous scores?

We treat pre-2019 Parker scores and post-2019 Vinous scores as a continuous source with the same weight of 10, reflecting the editorial continuity. Some methodology purists prefer to down-weight the transition era; Rankquant's v1 does not. This may be revised with a public version bump if the community consensus shifts.

By Ryan Siegal · Founder and Principal

Published 2026-04-23

The four scales you'll see

Parker 100-point (Wine Advocate, Vinous, Wine Spectator, Wine Enthusiast, Suckling, Dunnuck)

Nominal range 50-100; effective range 85-100. Parker explicitly designed the scale as a report-card: 90-95 "outstanding", 95-100 "extraordinary". In practice the sub-85 zone is nearly empty — wines scoring under 85 generally just don't get written up. That's a 15-point usable range pretending to be a 50-point range, which is why the scale has become so inflated over 40 years.

Jancis Robinson 20-point

Nominal range 12-20; effective range roughly 14-20. Jancis Robinson MW's house style is stricter: 18+ is rare, 19 is reserved for wines she considers excellent-for-the-ages, 20 almost never appears. Less inflation per unit on the scale, but lower coverage (only wines Jancis personally tastes).

Vivino / Cellar Tracker / retailer 1-5 stars

Nominal range 1-5; effective range 3.0-4.8 or so. Crowd platforms suffer the same rating-inflation problem as Amazon and Goodreads. Vivino averages around 3.8 across the whole catalog. A 4.5 is a strong wine; a 4.0 is modest; anything below 3.5 is notably flawed.

Rankquant normalized 1-5

By construction: 5 means top 5% of the peer set, 3 means peer-set average, 1 means genuinely underperforming. Not inflatable because the distribution is defined by the normalization itself, not by reviewer incentives.

Scale conversion table

The linear conversions Rankquant uses to rescale every source to a common 0-100 range for the weighted-mean step of the methodology:

Rescaling constants. Each source's native score is linearly transformed to 0-100 before the source-weighted mean step.
Parker 100-pt	Identity (no change). 89 → 89.
Jancis 20-pt	×5. 17 → 85.
Vivino 1-5	×20. 4.3 → 86.
CellarTracker 100-pt	Identity. 88 → 88.
Retailer 1-5	×20. 4.4 → 88.
Wine-Searcher community	Identity (100-pt). Low weight.

Rescaling constants. Each source's native score is linearly transformed to 0-100 before the source-weighted mean step.

Rough cross-walk between scores you can use in a store

Rough cross-walk for wines that receive scores from multiple sources. Applies only to wines actually rated; does not correct for inflation pressure differences between houses.
Parker 100	Rankquant 5.0 · Jancis 20 · Vivino 5.0
Parker 95	Rankquant 4.9+ · Jancis 19 · Vivino 4.6-4.8
Parker 93	Rankquant 4.7 · Jancis 18 · Vivino 4.4-4.6
Parker 90	Rankquant 3.8-4.2 · Jancis 17 · Vivino 4.2-4.4
Parker 88	Rankquant 3.2-3.6 · Jancis 16 · Vivino 4.0-4.2
Parker 86	Rankquant 2.5-3.0 · Jancis 14-15 · Vivino 3.7-4.0
Parker <85 or unrated	Rankquant ≤ 2.5 · Vivino < 3.7

Rough cross-walk for wines that receive scores from multiple sources. Applies only to wines actually rated; does not correct for inflation pressure differences between houses.

Practical consumer rules

Ignore any wine rated under 88 by a professional 100-pt source.The critic didn't think it was worth writing up highly; the lower score is often a euphemism for "ordinary".
A 90 is not "excellent". In Parker-era scoring a 90 is a competent wine, nothing more. The marketing industry has convinced consumers otherwise.
Cross-check before buying on a single score. If only one critic rates a wine, its score carries less signal. Rankquant explicitly handles this via Bayesian adjustment — thin-critic-coverage wines are pulled toward the peer-set prior.
Peer-set context is the real signal. A Parker 92 on a $25 Bordeaux is remarkable. A Parker 92 on a $200 Bordeaux is disappointing. The raw score is the same; the normalized-within-price-tier score is wildly different.
Prefer the normalized score when available.A Rankquant normalized 4.7 with published peer-set "2019 Napa Cabernet $40-$80, 213 wines" tells you more than a Wine Spectator 93 with no context.

Frequently asked questions

Is a 95 Parker score really better than a 92?+

Yes, but the gap is smaller than it looks because the effective scale compresses 15 points into 15 ranks. A 95 vs 92 is roughly the difference between top 10% and top 30% of wines rated by Parker — meaningful but not as dramatic as "95 vs 92" sounds in absolute terms.

Why does Rankquant rescale crowd ratings so aggressively?+

Because the math requires sources to be on a common scale before weighting. Vivino × 20 brings the crowd 1-5 onto a 20-100 effective range that the source-weighted mean can aggregate with 100-point professional scores. The rescaling is purely mechanical; it's not a claim that a Vivino 4.3 equals a Parker 86.

What about the 5-star "half-star" increments?+

Handled as decimals (4.5 → rescaled to 90, 4.3 → 86, etc.) before aggregation. No special treatment.

Do older Parker scores still count the same as recent Vinous scores?+

We treat pre-2019 Parker scores and post-2019 Vinous scores as a continuous source with the same weight of 10, reflecting the editorial continuity. Some methodology purists prefer to down-weight the transition era; Rankquant's v1 does not. This may be revised with a public version bump if the community consensus shifts.