The influence of signal probability during pretest practice on performance at a visual detection task was investigated in a 2 X 2 design, involving 2 levels of signal probability, "high" (p = .18) and "low" (p - .02), and 2 phases of the experiment, a practice period and a test session. The Ss practiced with the high signal probability detected significantly more signals than Ss practiced with the low signal probability (p < .01). The degree of within-session decrement varied with the different combinations of practice and test probabilities, being greatest in the high practice-low test probability case, and least in the low practice-low test probability case. The time of occurrence of the 1st signal in the test session was also found to affect the decrement function. These results support an expectancy theory of vigilance, and cast doubt on the validity of the conclusions drawn from earlier studies which have failed to control signal probability during pretest practice.