Memory improves when encoding and retrieval processes overlap. Here, we investigated how the neural bases of long-term memory encoding vary as a function of the degree to which functional processes engaged at study are engaged again at test. In an incidental learning paradigm, electrical brain activity was recorded from the scalps of healthy adults while they made size judgments on intermixed series of pictures and words. After a 1-hr delay, memory for the items was tested with a recognition task incorporating remember/know judgments. In different groups of participants, studied items were either probed in the same mode of presentation (word–word; picture–picture) or in the alternative mode of presentation (word–picture; picture–word). Activity over anterior scalp sites predicted later memory of words, irrespective of type of test probe. Encoding-related activity for pictures, by contrast, differed qualitatively depending on how an item was cued at test. When a picture was probed with a picture, activity over anterior scalp sites predicted encoding success. When a picture was probed with a word, encoding-related activity was instead maximal over posterior sites. Activity differed according to study–test congruency from around 100 msec after picture onset. These findings indicate that electrophysiological correlates of encoding are sensitive to the similarity between processes engaged at study and test. The time course supports a direct and not merely consequential role of encoding–retrieval overlap in encoding. However, because congruency only affected one type of stimulus material, encoding–retrieval overlap may not be a universal organizing principle of neural correlates of memory.