A survey of the results obtained during the past three decades in some of the most widely used tasks and paradigms in the experimental study of reasoning is presented. It is shown that, at first sight, human performance suffers from serious shortcomings. However, after the problems of communication between experimenter and subject are taken into account, which leads to clarify the subject's representation of the tasks, one observes a better performance, although still far from perfect. Current theories of reasoning, of which the two most prominent are very briefly outlined, agree in identifying the load in working memory as the main source of limitation in performance. Finally, a recent view on human rationality prompted by the foregoing results is described.