Recent prospective memory (PM) studies indicate that intentions are not always directly deactivated after completion, but that they entail aftereffects in terms of slower ongoing-task performance and/or commission errors on repeated (no-longer relevant) PM trials. In four experiments, we investigated whether aftereffects depend on the similarity between completed and currently performed PM-tasks. Aftereffects were reduced when PM-cues differed between the two PM-tasks (symbols vs. words) compared to when PM-cues belonged to the same category (symbols vs. symbols). This could be explained by the new dissimilar PM-task shifting spatial attention away from repeated PM-cues and, thus, attenuating processing of these cues. Moreover, a switch of the PM-response (to or within the manual modality) did not result in erroneous retrieval of the no-more-relevant PM-response (i.e., commission errors) but in erroneous retrieval of the currently relevant PM-response (i.e., false alarms). In addition, aftereffects vanished in conditions, in which participants did not perform a new PM-task. This finding indicates that forming a new PM-task set might be a prerequisite for aftereffects when the ongoing task changes between the two subsequent PM-tasks. Finally, we did not find evidence that forming a new, dissimilar PM-task representation led to overwriting of the completed intention representation, and thus to a change of the content or destabilization of its activation level.