In two experiments we explored whether participants would be able to use probabilistic cues to simplify perceptually demanding visual search in a task we call the retrieval guidance paradigm. On each trial a background cue appeared prior to (and during) the search task and the diagnosticity of the background cue(s) was manipulated to provide complete, partial, or non-diagnostic information regarding the target’s color on each trial. Only when participants were made aware of the possible relationship between the background cues and target features were they able to utilize the cue information for search. When participants were not made aware of the possible connection, they were only able to use target base rates. In the General Discussion we address how a recent computational model of hypothesis generation (HyGene, Thomas, et al., 2008), provides a useful framework for understanding how long-term memory, working memory, and attention coordinate in visual search.