A common finding in distracter interference (e.g., Flanker) tasks is that the difference in mean reaction time (RT) between incongruent and congruent trials-the congruency effect-is smaller when the previous trial was incongruent relative to congruent. Over the past 2 decades, 2 main accounts of this congruency sequence effect (CSE) have been proposed. One posits that the CSE indexes trial-by-trial adjustments of cognitive control, which are triggered by expectation, response conflict, negative affect, or response suppression. The other holds that the CSE indexes feature integration and/or contingency learning processes that are confounded with congruency sequence in most studies. In 3 online experiments involving over 450 participants, we observed CSEs without such confounds when 2 preconditions were met: (a) stimulus-response translation could be completed more rapidly for the distracter than for the target and (b) the distracter and target appeared at the same location. We also found that CSE magnitude did not vary consistently with the size of the congruency effect. These findings reveal that CSEs can be observed in the absence of feature integration and contingency learning confounds, but impose important new constraints on certain cognitive control accounts of this phenomenon.