Declarative memory is a central resource for reasoning processes. In line with the ACT-R theory, we assume that declarative memory is the basis for causal learning. Based on this assumption we conducted an experiment, showing that subjects’ confidence in causal predictions decreased if their causal knowledge is discredited. Moreover, confidence decreased not only for the causal knowledge that was discredited, but also for knowledge that was not at all manipulated. Additional to the experimental results, we present an ACT-R model that perfectly fits the data and provides an explanation for the empirical findings. Contextual change turns out to sufficiently explain the empirical data and the principle of our ACT-R model.