Economists have long been intrigued by an influential literature in psychology positing that monetary pay lowers performance on enjoyable tasks by crowding out agents' intrinsic interest in them. But typical experiments in this literature do not report a full set of performance metrics, which might reveal conflicting evidence on crowding out. Further, they may suffer from confounds. To evaluate these issues, we review over 100 prior tests and run a field experiment building on the canonical two-session test for crowding out wherein agents receive pay for an interesting activity in session one that is withdrawn unexpectedly in session two. We test whether pay harms performance using a comprehensive set of performance measures, and if so, whether unmet pay expectations might also contribute to this decline. Our results on output, productivity and quits are most consistent with a standard economics model than with a crowding out one. Additional, though more speculative, evidence suggests that unmet pay expectations may harm output quality.