Performance in choice reaction time tasks deteriorates when an irrelevant stimulus feature is associated with an incorrect response (conflict condition). Such interference effects are reduced under conditions of increased conflict-frequency. Although models of cognitive control account for this modulation in terms of conflict-related attentional focusing on the target stimulus dimension, it is possible that the effect reflects practice with specific stimulus ensembles or stimulus feature-response contingencies. Using an Eriksen flanker task, we deconfounded the frequency of conflict trials and the frequency of specific stimulus ensembles (i.e., target-flanker conjunctions). In Experiments 1 and 2, flanker interference varied inversely with the frequency of conflict trials, irrespective of practice with specific target-flanker conjunctions, thereby disputing a stimulus ensemble learning account. In Experiment 3, however, flanker interference was reduced for specific flanker stimuli which occurred predominantly in conflict trials. Taken together, the findings are consistent with flanker-specific attentional adjustment or associative flanker-response priming.