The effect of a safety signal on the magnitude of anticipatory anxiety was investigated using the fear-potentiated startle reflex paradigm in humans. The amplitude of the acoustic startle reflex was measured during the anticipation of unpleasant electric shocks ("threat") and during "safe" conditions. Threat and safe conditions were signaled by three different colored lights. Two lights signaled safe conditions (safe 1, safe 2) and the other light signaled the threat condition (threat). In phase I, the lights alternated, each presentation consisting of one colored light. In phase II, the lights were presented alone or in the two combinations of safe 1 (or safe 2) + threat and safe 1 + safe 2. In both phases, the contingency between the lights and the shock was explained to the subjects. It was emphasized that no shock could be administered when the safe 1 and threat light were simultaneously presented in phase II. Subjects' belief and understanding of the instructions were verified. In Phase I, startle was increased in the threat-alone compared to the safe-alone condition, reflecting increased anticipatory anxiety in the threat-alone condition. In phase II, startle in the safe + threat condition was smaller than in the threat-alone condition, but was larger than in the safe + threat. These results were interpreted as suggesting that the threat signal was still able to elicit anticipatory anxiety despite the fact that it was no longer associated with a threat.