An illusory correlation (IC) experiment examined the presence of a phobia-relevant covariation bias in the context of social anxiety. Low (n = 28) and high (n = 32) social anxious women were shown a series of slides comprising pictures of angry, happy and neutral faces which were randomly paired with either a shock, a siren or nothing. One half of the participants were shown women faces, whereas the other half were shown men faces. Participants indicated outcome expectancies on a trial by trial basis. After the experiment proper they estimated the contingencies of all slide/outcome combinations. Participants showed both an a priori and an a posteriori IC between angry faces and shock. This covariation bias was similar for men and women faces and independent of prior fear. The pattern of results is consistent with the idea that ICs arise from initial expectancies that survive extinction.