In this study, we examine the role of beliefs in conditional inference in two experiments, demonstrating a robust tendency for people to make fewer inferences from statements they disbelieve, regardless of logical validity. The main purpose of this study was to test whether participants are able to inhibit this belief effect where it constitutes a bias. This is the case when participants are specifically instructed to assume the truth of the premises. However, Experiment 1 showed that the effect is no less marked than when this instruction is given, than when it is not, although higher ability participants did show slightly less influence of belief (Experiment 2). Contrary to the findings with syllogistic reasoning, use of speeded tasks had no effect on the extent of the belief bias (both experiments), although it did considerably reduce the numbers of inferences that were drawn overall. These findings suggest that the belief bias in conditional inference is less open to volitional control than that associated with syllogistic reasoning.