Abstract There has been an increased interest in seasonal forecasting of the Arctic sea ice extent in recent years, in particular the minimum sea ice extent. Here, a dynamical mechanism, based on winter preconditioning, is found to explain a significant fraction of the variance in the anomaly of the September sea ice extent from the long-term linear trend. To this end, a Lagrangian trajectory model is used to backtrack the September sea ice edge to any time during the previous winter and quantify the amount of sea ice advection away from the Eurasian and Alaskan coastlines as well as the Fram Strait sea ice export. The late-winter anomalous sea ice drift away from the coastline is highly correlated with the following September sea ice extent minimum . It is found that the winter mean Fram Strait sea ice export anomaly is also correlated with the minimum sea ice extent the following summer . To develop a hindcast model of the September sea ice extent—which does not depend on a priori knowledge of the minimum sea ice extent—a synthetic ice edge initialized at the beginning of the melt season (1 June) is backtracked. It is found that using a multivariate regression model of the September sea ice extent anomaly based on ice export from the peripheral Arctic seas and Fram Strait ice export as predictors reduces the error by 38%. A hindcast model based on the mean December–April Arctic Oscillation index alone reduces the error by 24%.