Perception is created by an unconscious inference procedure that reflects the structure of the world (Helmholtz, 1910/1925). Usually multiple relevant signals, or cues, are involved in constructing a percept. This process is known as ‘cue combination’. Cue combination has some dynamic features of presumed adaptive value that cause changes to the way cues are combined during perception. Some of these dynamic features (e.g. recalibration and re-weighting of cues) have been widely recognized. Another dynamic feature, cue recruitment, might also be necessary: the perceptual system needs to recognize the relevance of a potential cue for certain perceptual attributes. If so, we would expect that Pavlovian conditioning can build associations in perception, because the only way to know the relevance of a new cue is through its co-occurrence with some pre-existing cues (Brunswik, 1947/1956; Smedslund, 1955). However, this type of learning-through-conditioning in visual perception has in the past been difficult to demonstrate experimentally (Fahle, 2002; Gibson & Gibson, 1955). Here we demonstrate that a novel, experimentally created contingency between signals caused changes in the visual appearance of a bistable stimulus (Haijiang et al., 2006). We further investigated the impact of the newly recruited cue on the perceptual inference system. Specifically, a first set of additional experiments shows that the new cue was integrated into the perceptual inferential procedure as a separate source of information from the pre-existing cues (Backus & Haijiang, 2007). A second set of additional experiments demonstrates that the learning system generalized the newly recruited cue to signals of different intensities from the training stimulus. We hypothesize that cue recruitment is an integral part of the mechanism that the perceptual system relies on to draw the signals correlation map upon which the procedure of perceptual inference operates. Cue recruitment, together with current models of cue combination, presents a more complete picture for the problem of cue utilization in the perceptual inference system.