This paper reports on an experiment that distinguishes between learning rules that produce substantively rational behavior in more situations and learning simple rules that do not. The experimental task is a simplification of many naturally occurring search problems. The experiment manipulates individual choice problems such that simple rules produce substantively rational behavior in one condition but not in subsequent ones. Choice data indicate subjects may learn to use more rational rules since subjects successfully transfer what they learn in one condition to another. However, verbal protocols indicate subjects learn to selectively use simple rules in order to make more substantively rational choices.