In this paper we present a model of three central aspects ofprobabilistic cognition: event prediction, feature formation,and attention allocation. While most models of probabilisticreasoning take a parameter estimation and error minimisationapproach (sometimes referred to as ‘predictive coding’, and of-ten described in terms of Bayesian updating), our model takesa contrasting frequentist hypothesis-testing approach. Thischoice is motivated by a series of recent results suggesting thatpeople’s probabilistic reasoning follows frequentist probabilitytheory. In simulation tests we demonstrate that this frequentistmodel, in which predictive features are formed by a process ofnull hypothesis significance testing, can give a successful ac-count of event prediction and attentional switching behaviour.