The present study tested predictions derived from Rachman's match-mismatch theory of fear. Students (N = 117) displaying claustrophobic fear made predictions of fear, panic and performance prior to exposure to a claustrophobic chamber. Subjects were then classified into groups (i.e. matchers, overpredictors, underpredictors) based on the congruence between their anticipated phobic reactions and their actual reactions during exposure. The effects of match-mismatch status on predictions of fear, panic and performance on a subsequent exposure trial were examined. Results provided further evidence that phobic Ss are frequently inaccurate in predicting their phobic reactions during exposure. Consistent with previous findings, Ss who overpredicted fear, panic or performance tended to lower their predictions on a subsequent exposure trial, whereas Ss who underpredicted fear, panic or performance showed heightened predictions upon subsequent exposure. Contrary to previous findings, Ss were more likely to underpredict rather than overpredict their phobic reactions. Our findings are in accord with those of Rachman indicating that the magnitude of mismatch between predicted and actual fear/panic is more influential in determining subsequent predictions of fear/panic than it is in determining Ss' actual fear/panic. Moreover, this asymmetry was observed when the mismatch paradigm was applied to phobic behavior (i.e. tendency to escape).