Football Predictions for Today: Premier League and Weekend Picks for September 2026

Each row is one fixture. The bar in the middle splits the three results, a home win, a draw and an away win: by how likely the model thinks each is, so a bar that runs mostly dark is a strong home favourite, and a bar split evenly across the three is a genuine coin-toss. The percentages under it always add up to 100.

Likely score is the single most probable exact result, not an average: plenty of matches finish on a scoreline the model never flags, because football is low-scoring and luck plays a real part. Over 2.5 is the chance of at least three goals in the game, the number most punters reach for first when they think a match will be open. Read the three together and you get a shape for the fixture before you have looked at a single price.

The Premier League gets the deepest data, so our Premier League predictions are our most confident. Twenty clubs, thousands of past matches: the model has plenty to learn from, and the markets are liquid enough to hold value.

Every fixture of the round is priced in the board above, the probabilities, the likely score, the goals line. The strongest calls graduate to our premier league betting tips, where each selection carries a market, a price and the reasoning behind it. If you want the forecast, read it here; if you want our staked premier league betting tips, follow that link. Forecast here, bet there.

The engine is simple in principle. Estimate the goals, price the markets, compare to the bookmaker. Where our price is shorter than theirs, there is value, and that is the bet.

Underneath, it is a goals model. We take every result from the last full Premier League season and work out how many goals each club scores and concedes relative to the league average: separately at home and away, because home advantage is real and worth roughly a third of a goal a game. That gives every team an attacking rating and a defensive rating. Newly promoted sides have no top-flight record to learn from, so they start on a standard promoted-club baseline until results roll in and the numbers correct themselves.

For any fixture, those ratings combine into an expected-goals figure for each side. We run that through a Poisson distribution: the standard way to model how often a low, random event like a goal happens: to get the probability of every plausible scoreline. Add up the scorelines where the home team wins and you have the home probability; sum the ones with three or more goals and you have the over-2.5 figure; the single most probable cell is the likely score you see on the board.

It is not magic, and it will not win every week. Football is noisy by nature, a deflection, a red card or a missed penalty can bend a game the model rated as one-sided. But over a full season, backing value beats backing favourites on instinct. We show the working so you can judge it yourself, rather than take our word for it, and the whole board rebuilds automatically as each matchweek's fixtures come round.