Pseudocontingencies are inferences of correlations between variables, like two options and two outcomes, drawn on the basis of their skewed base rates covarying across a third variable (e.g., two contexts). Here, we investigated the effect of pseudocontingency inference on choice behavior. When choices between two options are not based on the actual contingency between options and outcomes, but instead on a pseudocontingency, the latter may override the existing contingency, resulting in potentially suboptimal choice behavior. Whereas research has mainly focused on investigating the pseudocontingency effect by presentation of predetermined learning trials, we examined the role of free information sampling for the pseudocontingency effect as compared with predetermined learning. Experiment 1 replicated previous findings of a pseudocontingency effect in choice behavior. In Experiment 2, we compared predetermined information and free information sampling in a bivariate decision scenario with only two options and two outcomes. Experiments 3 and 4 aimed at investigating the inference of a pseudocontingency when sampling information by context or by context and option in the trivariate scenario. The results revealed an asymmetry between positive contexts with predominantly gains and negative contexts with predominantly losses. Within a negative context we found no differences between options, neither during information sampling nor for subsequent choices. Within the positive context, when information sampling was self-determined, participants sampled skewed base rates of options and preferred the predominant option. The findings underline the influence of self-determined information sampling on the pseudocontingency effect on choice behavior.