Summary. Using the British Household Panel Survey 1990-2007, we estimate a discrete duration model of interregional migration in Great Britain. By exploiting retrospective information on residency we control for late entry as well as unobserved heterogeneity. We find considerable duration dependence in region of residence in the raw data, much of which disappears when controlling for observable and unobservable differences between individuals. We illustrate that the usual way of modelling the effect of regional variables on the propensity to migrate gives insignificant and unrobust results and we introduce a new competing risks technique. © 2011 Royal Statistical Society.