We studied the effects of expectation and repetition on searching for singletons in very brief displays. In Experiment 1, we found that when a given feature defined the singleton for a whole block--so that in every trial the subject could expect a particular target feature--search accuracy was significantly higher than when the feature setting was randomly redetermined from trial to trial. However, an unexpected repetition triggered almost no advantage. In Experiment 2, we found no advantage for expected alternation. Experiments 3 and 4 demonstrated that there is little or no advantage for perception in conditions allowing only for target-feature facilitation or distractor-feature inhibition. We propose that in singleton search a division of feature space facilitates detection, and that this division works best under conditions of expected repetition. Experiment 5 replicated Experiment 1, but in Experiment 5 we examined response times and long display exposures. Results suggested that previous findings of singleton priming reflect mainly postperceptual factors.