This paper investigates the effect of three important variables on signal use, theoretically and experimentally. We present a simple model and a corresponding experiment that investigates the combined effects of uncertainty about action, signal reliability and signal cost. Our experiment uses techniques drawn from the psychology laboratory to test the behavior of captive blue jays Cyanocitta cristata solving a simple feeding problem. In the experiment, individual jays must choose which of two keys to peck; one key leads to food, but the other does not. In addition, the jays can choose to see a signal that provides information about which key provides a food reward. We find that these three variables interact to determine signal use crudely as our model predicts. Specifically, we find maximal signal use when uncertainty is high, signals are reliable and signal cost is low, but the effects of these variables on signal use do not combine additively. A change in any single variable can abolish signal use. While our results agree qualitatively with our simple model, we also find - in agreement with previous studies - that subjects show a bias against signal use. Specifically, they tend to ignore signals and rely on prior information about the correct action in 'intermediate' conditions. These results are important for two reasons. First, they replicate earlier results from our laboratory using a different preparation. Second, they highlight the central interaction between uncertainty and reliability that determines the value of signals for receivers. This is significant because game theoretical models of signaling emphasize the problem of 'reliability,' but they pay little attention to the fundamental interaction between reliability and uncertainty.