In this study we examine whether learning the statistical structure of successive stimuli improves performance on the flanker task. In three experiments, sequences of flanker stimuli were segregated into pairs, with the second stimulus contingent on the first. RTs were lower for the second stimulus if its congruence tended to match the congruence of the preceding stimulus (Experiment 1). In contrast, performance was not enhanced if the congruence of the second stimuli tended to mismatch the congruence of the first stimulus (Experiment 2). The distinction appears to result from the difficulty of learning mismatch contingencies (Experiment 3). Our results suggest that implicit learning of inter-stimulus relationships can facilitate cognitive control of attention.