Safety signal learning in human subjects was investigated using a conditioned inhibition procedure with a shock unconditioned stimulus (US). The conditioned stimulus (CS) duration, and thus the CS-US interval, varied randomly from 20 to 50 s. Conditioning was assessed by change in tonic skin conductance level (SCL). Experiment 1 demonstrated reliable learning of a conditioned inhibition (A+/AB-) discrimination, in terms of both self-reported shock expectancy and change in SCL, in subjects who were able to report correctly the stimulus contingency. There was, however, no evidence of transfer of inhibitory properties from the safety signal B to a separately trained excitor. Experiment 2 compared two groups in which the safety signal was either fear-relevant (picture of snake or spider) or fear-irrelevant (flower or mushroom). As in previous research, there was no effect of fear-relevance on safety-signal learning. The results are discussed in terms of preparedness theory and excitatory conditioning with fear-relevant stimuli.