It is hypothesized that the transition from unfamiliar problems to familiar, well-learned problems coincides with the transition from a model-based style of reasoning to a rule-based style of reasoning; model-based style of reasoning helps understanding the problem structure, but can overload working memory when the number of models required increases; rule-based style of reasoning avoids cognitive overloading, at the cost of making individuals liable to mechanization errors. In Experiment 1, the number of models required to respond to a verification task affected response latencies with unfamiliar problems, but not with familiar problems, supporting the initial hypothesis. In Experiment 2, participants were prone to mechanization errors when confronted with slightly modified problems in the late stages of the experiment, supporting the hypothesis that they had developed a reasoning rule in the early stages and were blindly applying it. The findings suggest that model-based reasoning and rule-based reasoning serve different purposes and have different costs and benefits, are both available to human reasoners, and familiarization with a problem may induce the transition from the former to the latter. The findings also suggest that mechanization of reasoning may be the first step along a gradient of decreasing cognitive load, whose end-point is automatization, as discussed in theories of automatization of information processing.