We conducted six treatments of a standard moral hazard experiment with hidden action. The behavior in all treatments and periods was inconsistent with established agency theory. In the early periods, behavior differed significantly between treatments. This difference largely vanished in the final periods. We used logit agent quantal response equilibrium (LAQRE) as a device to grasp boundedly rational behavior and found the following: (1) LAQRE predictions are much closer to subjects' behavior in the laboratory; (2) LAQRE probabilities and experimental behavior show remarkably similar patterns; and (3) including social preferences in LAQRE does not better explain the experimental data; (4) LAQRE cannot explain the contract offers of some players who seem to choose some focal contract parameters.