Standard theoretical search models assume that agents behave as if they search optimally. This paper reports the experimental results of a test of whether agents maximize their payouts by optimally searching for the best of n candidates in what has been called the "Secretary Problem." The novelty of this design is that the optimal search rule for a given subject is invariant to that subjects' risk attitude. This is made possible by implementing a binary payout schedule in which stopping at the best candidate pays a positive amount and stopping at any other candidate pays nothing. On average, subjects chose the best candidate slightly less than a fictitious optimal searcher, while the El-Gamal and Grether (1995) estimation procedure suggests that subjects are either using a sub-optimal rule or just randomly guessing.