Abstract This paper describes three experiments examining predictions from the expectancy bias model of selective associations (Davey, 1995). In a simulated ‘threat’ conditioning procedure, Experiment 1 showed that UCS expectancy following both ontogenetic and phylogenetic CSs was significantly predicted by: (1) ratings of the dangerousness of the CS, perceptions of CS-UCS similarity, and level of prior fear to the CS; and (2) ratings of CS-UCS similarity on the dimensions of valence, arousal and anxiety. Experiment 2 used a covariation assessment procedure which confirmed the findings of Experiment 1, and also showed that both phylogenetic and ontogenetic fear-relevant CSs exhibited both a priori and a posteriori covariation biases. Experiment 3 found that Ss high and low in fear to a fear-relevant CS exhibited a significant a priori UCS expectancy bias, but this bias was significantly larger in high fear Ss. Only high fear Ss exhibited an a posteriori covariation bias. These results are consistent with predictions from the expectancy bias model.