Verbal working memory (WM) has been assumed to involve two different systems of maintenance, a phonological loop and a central attentional system. Though the capacity estimate for letters of each of these systems is about four, the maximum number of letters that individuals are able to immediately recall, a measure known as simple span, is not about eight but six. We tested the hypothesis that, unaware of the dual structure of their verbal WM, individuals underuse it by trying to verbally rehearse too many items. In order to maximize the use of verbal WM, we designed a new procedure called the maxispan procedure. When performing an immediate serial recall task, participants were invited to cumulatively rehearse a limited number of letters, and to keep rehearsing these letters until the end of the presentation of the list in such a way that the following letters can no longer enter the phonological loop and must be stored in the attentional system. As we expected, in three successive experiments, the maxispan procedure resulted in a dramatic increase in spans compared with the traditional simple span procedure, with spans approaching eight when the to-be-rehearsed letters were presented auditorily and the following letters visually. These results indicate that simple spans, which have been used for more than a century in intelligence tests and are assumed to measure the capacity of short-term memory, actually reflect the complex interplay between different structures and cognitive processes.