Six experiments examined the determinants of the numeral advantage effect: the finding that memory span for Arabic numerals (1, 2, 3, etc.) is greater than for digit words (one, two, three, etc.). The speed of item identification for numeral and digit words was unrelated to memory span for the same items and a larger memory span for numerals persisted under concurrent random generation (Experiment 1). The numeral advantage, however, was abolished when the items were presented in random locations within an invisible 3x3 grid (Experiment 2) and in locations on a horizontal plane that ran contrary to the natural direction of reading (Experiment 3). When the items were presented in the same location, a disruption of the spatial component of visuo-spatial working memory eliminated the numeral advantage (Experiment 4), whereas interference with the visual component of the system did not (Experiment 5). When the items were spatially distributed in a 3x3 matrix, however, neither visual nor spatial interference abolished the effect (Experiment 6). Taken together, these findings suggest that the numeral advantage effect is mediated by discrete components in visuo-spatial working memory dedicated to the temporary storage and renewal of visual codes and questions the assumption that the underlying mechanisms in immediate, visual serial recall are equivalent between stimulus categories.