This ERP study aimed to investigate the underlying mechanisms of directed forgetting of emotion and neutral words. Cognitive theories suggest that multiple processes underlie the effects of directed forgetting, including differential encoding of to-be-remembered (TBR) and to-be-forgotten (TBF) items, and selective inhibition of TBF items during retrieval. It has furthermore been demonstrated that the emotional nature of test items often interferes with memory performance, possibly due to increased arousal and/or semantic cohesiveness. Participants (N=17) were presented with 50 negative/anxious words and 50 neutral words. The colour in which the words were presented indicated whether they needed to be remembered or forgotten. In the test phase, participants were required to indicate recognition of the previously presented words, regardless of encoding instruction. Recognition performance was better and quicker for TBR items than TBF items. There was also a significant recognition bias (Br) for emotion items. ERPs during encoding showed a prolonged fronto-central positivity (400-600 ms post-stimulus) for the emotional items compared to the neutral items. A similar ERP effect was found during the test phase, although it had a slightly earlier onset and a more widespread scalp distribution. There was no apparent ERP effect of remember or forget instruction during encoding, suggesting that these items were not differentially encoded. Finally, the parietal ERP old/new effect was present for the TBR items but not for the TBF items, suggesting that directed forgetting effects in this study/test paradigm were primarily the result of mechanisms acting during the retrieval phase.