Scalp-recorded ERPs were employed in four experiments to explore the strategic control of recollection in a recognition exclusion task. For each experiment, the study phase comprised two study lists. The test required the endorsement of 'targets' from study list 2 and the rejection of 'non-targets' from study list 1 and 'new' items. Experiments 1-3 showed that the ERP signature for recollection, the 'left parietal old/new effect', was elicited by correctly rejected non-targets only when memory for targets was poor (cf. Herron & Rugg, 2003a). These findings support the proposal that, when there is good memory for targets, the adoption of a 'retrieval orientation' allows test cues to selectively probe memory for target source information. However, when target and non-target study tasks were identical (Experiment 4), left parietal effects were additionally found for non-targets despite high target accuracy. This indicates that the degree of similarity between target and non-target study contexts moderates the extent to which it is possible to focus retrieval attempts exclusively on target memories. However, when target and non-target study contexts were partially distinct (Experiment 4), a late posterior negativity was seen to attenuate an emerging left parietal effect for non-targets suggesting that non-target recollection was not completely precluded in this condition. Furthermore, the magnitude of the late posterior negativity was observed to correlate with the amount of contextual information associated with each item type that was irrelevant to the task demands. It was proposed that this negativity reflects processes acting upon a mismatch between the targeted memory representation and the contextual details that are actually retrieved. When such mismatches occur, an additional strategic control process, 'attentional suppression', may provide an account of how target memories can be successfully isolated from among competing alternatives.