Standard economic models with complete information predict a positive, monotonic relationship between pay and performance. This prediction does not always hold in experimental tests: offering a small payment may result in lower performance than not offering any payment. We test experimentally two main explanations that have been put forward for this result: the "incomplete contract" hypothesis views the payment rule as a signal given to subjects on purpose of the activity. The "informed principal" hypothesis views it as a signal concerning the characteristics of the agent or of the task. The incomplete contract view appears to offer the best overall explanation for our results. We also find that high-powered monetary incentives do not "crowd out" intrinsic motivation, but may elicit "too much" effort when intrinsic motivation is very high.