Fantasy Premier League player projections

Projected points for every Premier League player, rebuilt as the data moves. Palmer leads gameweek 6 on 5.90 points.

Rebuilt .

Highest projected scorers, gameweek 6

PlayerClubPosPricexPtsCeilingHaul
PalmerChelseaMID£9.7m5.9013.619%
B.FernandesManchester UnitedMID£12.0m5.7313.621%
SakaArsenalMID£9.5m5.6813.019%
João PedroChelseaFWD£7.8m5.2313.016%
HaalandManchester CityFWD£15.6m5.1410.613%
GroßBrighton & Hove AlbionMID£5.8m5.0510.415%
GabrielArsenalDEF£8.0m4.989.58%
RogersChelseaMID£7.7m4.9310.412%
BarnesNewcastle UnitedMID£6.1m4.8010.411%
TarkowskiEvertonDEF£6.1m4.759.57%
IsakLiverpoolFWD£9.1m4.7210.611%
ThiawNewcastle UnitedDEF£5.0m4.699.08%
BotmanNewcastle UnitedDEF£5.0m4.609.56%
Gibbs-WhiteNottingham ForestMID£8.0m4.5910.412%
RiceArsenalMID£7.4m4.559.39%
MbeumoManchester UnitedMID£7.9m4.4910.411%
ThiagoBrentfordFWD£7.8m4.4910.610%
GonzaloFulhamFWD£6.0m4.4810.612%
BarryEvertonFWD£5.6m4.469.510%
HallNewcastle UnitedDEF£5.2m4.328.67%
Projected points for gameweek 6. Ceiling is the 90th percentile of the simulated distribution; haul is the chance of ten points or more.

Every club

By position, over the whole run

This gameweek

How well it has done

1.05average error, points per player per gameweek
1.023calibration slope (1.000 is perfect)
25,750player-gameweeks graded
667players projected this run

Every projection published here has been graded the same way: predict, wait, compare. The calibration slope says whether the numbers are the right size — a model that says six and delivers four is no use even if it ranks players perfectly. Across the same test the projections correlate 0.57 with what players actually scored, which is what one gameweek of football looks like: the ranking is informative, a single week is mostly noise.

Open the app

More projections

Best goalkeepersBest defendersBest midfieldersBest forwardsGameweek 6

About these figures

These figures are produced automatically from our own projection model and rebuilt as new data arrives — team news, results, prices and fixtures. Last rebuilt .

These are averages over thousands of simulated gameweeks, not predictions of what will happen: a single week can land a long way from the average, in either direction. The model reads published data only — it has no access to press conferences or leaked team sheets. Graded by walk-forward testing across a completed season, its average error is 1.05 points per player per gameweek over 25,750 graded player-gameweeks. That grading is blind to team news by design: each week is rebuilt from match results alone, so every player in it is scored as though he was fit and available.

The method, and where the model is measurably wrong, are set out in full on how it works.