We experimentally study the Gale and Shapley, 1962 mechanism, which is utilized in a wide set of applications, most prominently the National Resident Matching Program (NRMP). Several insights come out of our analysis. First, only 48% of our observed outcomes are stable, and among those a large majority culminate at the receiver-optimal stable matching. Second, receivers rarely truncate their true preferences: it is the proposers who do not make offers in order of their preference, frequently skipping potential partners. Third, market characteristics affect behavior: both the cardinal representation and core size influence whether laboratory outcomes are stable. We conclude by using our controlled results and a behavioral model to shed light on a number of stylized facts we derive from new NRMP survey and outcome data, and to explain the small cores previously documented for the NRMP.