OddsLens
Data desk · Draw bias · Published 26 July 2026

Goodwood draw bias: what 122 races of data say

The short answer: across 122 Goodwood flat races with fields of eight or more (March 2025 – July 2026, our own results database), horses drawn in the high third of the stalls won 22% of races against a 33% fair share of runners. Low-drawn horses won 41% against a 36% share. Wide is the wrong place to be on the Sussex Downs — with caveats we'll spell out rather than bury.

With Glorious Goodwood starting Tuesday, here are the numbers, the method, and — because one course proves nothing — the same measurement run on Chester, York and Ascot so you can see the method finds bias in both directions, and finds nothing where nothing exists.

Goodwood, by trip

We compare each stall third's share of winners against its share of runners — if the draw didn't matter, the two would match.

TripDraw thirdWonFair shareRaces
Beyond a mile (round course)Low38.2%36.1%102
Middle37.3%30.5%
High24.5%33.4%
Up to a mileLow55.0%36.3%20
Middle35.0%30.1%
High10.0%33.7%

On the round course the story is less "low good" than "wide bad": low and middle both out-win their share, while high-third draws give up nearly nine percentage points. That fits how the track rides — tight turns, a camber that hangs horses out, and ground lost on every bend for anything caught wide.

The sprint sample is 20 races. Eleven low-third winners from twenty is a striking lean, and it agrees with the round-course pattern — but twenty races is twenty races. Treat it as a lean, not a law. We'd rather show you a small sample honestly than a big claim dishonestly.

The method, exactly

Flat races only, fields of 8+, March 2025 – July 2026, from the results database behind our public model record. Each horse's stall is scaled to its field (lowest third / middle / highest third of the stalls actually loaded that day), so a draw of 8 counts as high in a 10-runner race and middle in a 20-runner one. The "fair share" column is each third's share of runners, not a flat 33.3% — thirds don't split evenly in every field size, and pretending they do flatters the result. No going filter, no class filter, no cherry-picking of "qualifying" races.

Does the method hold up elsewhere?

Three checks. First, Chester — racing's most famous bias, and it shows:

Course & tripLow third wonFair shareHigh third wonFair shareRaces
Chester, beyond a mile46.3%37.4%26.3%33.1%95
York, beyond a mile44.7%35.8%28.5%33.5%123
Ascot, sprints23.5%35.2%45.1%33.4%51

Second, York's round-course races lean low almost as hard as Chester's. Third — and this is the important one — Ascot sprints lean the other way: high draws won 45% against a 33% share in our window. A method that only ever "found" low bias would be suspect. This one finds high bias where high bias lives, and finds roughly nothing at several courses we checked (Ascot beyond a mile, for one, runs close to fair). Draw bias is course-specific, direction-specific and trip-specific — anyone quoting one universal rule is guessing.

Before you back every low draw at the festival

Two honest problems. First, the market knows about draw bias too — a well-drawn horse at Goodwood is priced like a well-drawn horse at Goodwood. The bias being real doesn't make backing it profitable; by the off, most of it is in the price. If there's an edge, it's in the mornings, before the market fully settles — the same closing-line logic that runs through everything we publish. Second, stalls positioning moves (stands' side, centre, far side) and going changes bias — a soft-ground week can redraw the map. Our numbers are the average of the last 17 months, not a promise about Tuesday.

During the festival our model's pick in every Goodwood race — logged before the off, settled in public, losses left in — is on the festival hub, and every pick ever made is on the record. The draw is one input among the 40+ the model weighs.

The Goodwood festival hub →