Finding an object in our environment is an important human ability that also represents a critical component of human foraging behavior. One type of information that aids efficient large-scale search is the likelihood of the object being in one location over another. In this study we investigated the conditions under which individuals respond to this likelihood, and the reference frames in which this information is coded, using a novel, large-scale environmental search paradigm. Participants searched an array of locations, on the floor of a room, for a hidden target by pressing switches at each location. We manipulated the probability of the target being at a particular set of locations. Participants reliably learned target likelihoods when the possible search locations were kept constant throughout the experiment and the starting location was fixed. There was no evidence of such learning when room-based and body-based reference frames were dissociated. However, when this was combined with a more salient perceptual landmark, an allocentric cuing effect was observed. These data suggest that the encoding of this type of statistical contingency depends on the combination of spatial cues. (Contains 5 figures and 1 footnote.)