The purpose of this study was to determine whether Anderson’s ACT-R fan model could account for the fan effect under more complex conditions. Specifically, overlapping datasets were used so that related facts learned in one experiment could potentially affect the fan in other experiments. The study of the overlapping datasets made it possible to study an inference task by combining facts learned in separate experiments. Four experiments with human subjects were carried out and human performance was compared to predictions from Anderson’s ACT-R fan model. The results showed ACT-R fan model could be used as a basic building block for explaining complex fan tasks (some high-level cognitive tasks); ACT-R fan model with no adjustments to the parameters provided a reasonable account for human performance across all of the experiments. The results suggest that recently learned related facts have an effect on the fan (Experiment 2). But, it was found that the related facts learned ten months earlier showed no interference due to fan but there was a main learning effect which affected reaction times (Experiment 4). In terms of relational inferences from overlapping datasets, the results indicated a dual retrieval process with additional search process is more consistent with Anderson’s fan model than with Radvansky’s mental models approach or a parallel retrieval approach. Both Radvansky’s mental models approach and a parallel retrieval approach predicted a single retrieval process (Experiment 3).