The comparative testing of models of cognitive judgements has previously been done bypost hocparameter estimation on data gathered without feedback. If on any trial within an experiment responses are tested against competing models, and feedback is provided to the subject on his partial consistency with a currently best‐fitting model, then under some conditions behavioural and parameter convergence may be stochastically achieved. An algorithm SIMFBS with parameter convergence potential is described and its performance examined on up to five models of similarity, with various forms of memory built into the algorith