Past research and theory on causal inference has not explored the effects of varying the reliability of information. In two experiments, subjects judged either ability (given performance and effort information) or effort (given performance and ability information) where both the reliability and value of the given information varied. Individual differences were found in the judged relationship between ability and effort. Some judged ability and effort to be positively related, whereas others, judged ability and effort to be negatively related. These groups also differed in the way information reliability influenced their judgments, The positive group showed effects that agree with either an averaging or correlational model: the higher the reliability of one type of information, the less the effect of the other type of information. For the negative group, an increase in the reliability of one type of information actually increased the effect of the other type of information, a result that is inconsistent with the averaging model. Both an expectancy-contrast model and a correlational model can account for the results of the negative group. The different effects of information reliability for the two groups can be interpreted as evidence of two different inference processes. The results show the flexibility of human judgment strategies and the need for research considering variables that influence strategy use. The present research deals with the issue of information credibility in causal inference, specifically with how the reliability of information affects inferences of ability and effort. Although there are many social psychological theories of causal inference (Kelley, 1972, 1973; Reeder & Brewer, 1979), there has been little attempt to formulate a theoretical representation of the way information credibility influences causal inferences. The present research tests alternative representations of the information integration process for causal inferences. Because models of information integration can be regarded as formal representations of causal schemata (Surber, 1981b), the present study can be seen as part of a research program aimed at specifying the variables that influence when different infor