When people are shown a series of trials on which a cue and outcome are objectively unrelated (a null contingency), they often judge the causal strength of the cue to be positive, a phenomenon referred to as a causal illusion. This task has been taken as a laboratory model of the development of false causal beliefs. However, in three experiments, we found that participants reliably provided a positive causal rating for the cue prior to any trials having been experienced. Over null contingency trials, participants partially corrected their predictions of the outcome, but maintained their positive causal beliefs, especially with high (75 %) cue and outcome densities. An attempt to reduce the positive prior belief by scenario instructions was unsuccessful. Pre-training with a genuine positive or negative contingency modulated causal ratings in the expected direction, but did not alter the final causal bias. These results suggest that causal illusions may not be acquired but represent a failure to correct an initial positive prior belief.