Divided attention during encoding typically produces marked reductions in later memory. The attentional boost effect (ABE) is a surprising variation on this phenomenon. In this paradigm, each study stimulus (e.g., a word) is presented along with a target or a distractor (e.g., different colored circles) in a detection task. Later memory is better for stimuli co-occurring with targets. The present experiments indicate that the ABE arises during an early phase of memory encoding that involves initial stimulus perception and comprehension rather than at a later phase entailing controlled, elaborative rehearsal. Experiment 1 demonstrated that the ABE was robust at a short study duration (700 ms) and did not increase with increasing study trial durations (1,500 ms and 4,000 ms). Furthermore, the target condition is boosted to the level of memory performance in a full-attention condition for the short duration but not the long duration. Both results followed from the early-phase account. This account also predicts that for very short study times (limiting the influence of late-phase controlled encoding and thus minimizing the usual negative effect of divided attention), the target condition will produce better memory than will the full-attention condition. Experiment 2 used a study time of 400 ms and found that words presented with targets lead to greater recognition accuracy than do either words presented with distractors or words in the full-attention condition. Consistent with the early-phase account, a divided attention condition actually produced superior memory than did the full-attention condition, a very unusual but theoretically predicted result.