The aim of the current study was to examine whether depth of encoding influences attentional capture by recently attended objects. In Experiment 1, participants first had to judge whether a word referred to a living or a nonliving thing (deep encoding condition) or whether the word was written in lower- or uppercase (shallow encoding condition), and they then had to identify a digit displayed midway in a rapid serial visual presentation (RSVP) stream of 8 pictures. A picture corresponding to the previously processed word was presented either before or after the target digit. The results showed that this picture captured attention, thus resulting in an attentional blink for identification of a target digit, in the deep encoding condition but not in the shallow encoding condition. In Experiment 2, this capture effect was found to be abolished when an additional working-memory (WM) task was performed directly after the word-judgment task, suggesting that the capture effect stemmed from residual WM activation that could be erased by means of a secondary WM task. Taken together, these results suggest that deep and shallow encoding result in different degrees of WM activation, which in turn influences the likelihood of memory-driven attentional capture.