This paper points to the importance of prior beliefs in understanding the gap between decisions from experience and decisions from description. It puts forward a two-stage account that effectively incorporates prior beliefs into the examination of decisions from experience. The two-stage account assumes that (1) the subjective probabilities are estimated in a Bayesian manner, combining prior beliefs with observations, and (2) the estimated probabilities are transformed by probability weighting. The first stage provides a Bayesian explanation for the commonly found overestimation of infrequent outcomes, and an empirically appealing way to deal with always – or never – observed outcomes. A source dependent probability weighting in the second stage captures deviations from Bayesian rationality under experienced uncertainty. The two-stage model is tested by reanalyzing the data sets in Glöckner et al. (2016), as well as the famous Technion Prediction Competition data set of Erev et al. (2010). Model comparisons using BIC scores indicate that the two-stage model performs better than the single stage model approximating subjective probabilities with observed relative frequencies. The estimation results show that the two-stage model can accommodate both the classic and the reversed description – experience gap.