We outline a simple way of representing sets of non-normative judgements that makes them look as similar as possible to normative ones. This representation allows us to view certain types of non-normative judgments, such as conjunction fallacies, as arising from a misestimation of the correlation between events, that might arise when decision-makers have no prior information about the frequency of co-occurrence. We suggest that decision-makers use the perceived similarity between events to make inferences about correlation, and we describe the results of an experiment showing that judged correlation and violations of independence in probabilistic judgments are strongly influenced by the perceived similarity between events.