Long-term memory (LTM) is enhanced for emotional information, but the influence of stimulus emotionality on short-term memory (STM) is less clear. We examined the electrophysiological correlates of improved visual STM for emotional face identity, focusing on the P1, N170, P3b and N250r event-related potential (ERP) components. These correlates are taken to indicate which memory processing stages and cognitive processes contribute to the improved STM for emotional face identity. In the encoding phase, one or three angry, happy or neutral faces were presented for 2 s, resulting in a memory load of one or three. The subsequent 1-s retention phase was followed by a 2-s retrieval phase, in which participants indicated whether a probe face had been present or not during encoding. Memory performance was superior for angry and happy faces over neutral faces at load three. None of the ERP components during encoding were affected by facial expression. During retrieval, the early P3b was decreased for emotional compared to neutral faces, which presumably reflects greater resource allocation to the maintenance of the emotional faces. Furthermore, the N250r during retrieval was increased for emotional compared to neutral faces, reflecting an enhanced repetition effect for emotional faces. These findings suggest that enhanced visual STM for emotional faces arises from improved maintenance and from improved detection of face repetition at retrieval. (Contains 5 figures and 2 tables.)