Wagner’s fully elaborated theory of learning (e.g., Vogel, Ponce, & Wagner, 2019) was founded on an initial analysis of the mechanisms responsible for habituation (Wagner, 1976, 1979). Central to its explanation of long-term habituation was the proposal that a predicted stimulus, one signaled by some other event as a consequence of associative learning, would be less effective at activating its central representation. We review evidence (from studies of the role of context in habituation and latent inhibition, of preexposure to the event to be used as an unconditioned stimulus in conditioning, and of conditioned diminution effects) taken to support this explanation. We argue that the evidence is less than convincing and consider instead an alternative account that interprets habituation as reflecting a reduction in the effective salience of a stimulus that is determined by a learning process akin to extinction, in which the critical factor is that the stimulus is presented followed by no consequences. The application of this account to the phenomena dealt with by Wagner’s model is considered and further implications are discussed.