The effects of word frequency on memory span were explored using the "up-and-down" method. Mean spans were greatest when the sequences were of all high-frequency words (5.82), and smallest when they were of all low-frequency words (4.24). For mixed-frequency sequences, mean spans were greater when the high-frequency words were presented before the low-frequency words (5.19) than when the low-frequency words came first (4.65). The findings are discussed in terms of the primary-secondary memory distinction worked out for single-trial free recall, and the logic of this distinction is used to argue against attributing span performance to a simple unitary process.The concept of memory span is as old as the experimental study of memory itself. Thus, in the very first such investigation, Ebbinghaus (l885/1964) found that he could almost always reproduce a series of up to seven items after just a single presentation, whereas for longer series correct reproduction usually required many presentations. Shortly after publication of Ebbinghaus' monograph, it was shown that the immediate memory span, or "span of prehension," was a valid measure of mental capacity (Galton, 1887; Jacobs, 1887), and it soon found its place in clinical diagnosis and intelligence testing.Despite its splendid history, the concept of memory span has been subjected to remarkably little critical analysis. Today we know hardly any more about memory span, about the extraordinary effect of increasing list length over a comparatively short range, than Ebbinghaus knew. This neglect is made the more striking by the intensive current interest in the immediate serial recall of supraspan lists. No doubt the primary reason for this discrepancy lies in the difference in the readiness with which span and supraspan recall submit to analysis. In particular, the items of supra span lists can be classified according to whether or not they are recalled, thereby allowing serial position functions to be drawn, analyzed, and given a theoretical interpretation. With span recall, on the other hand, we lack suitable techniques for revealing any influence of serial position. The purpose of this paper is to suggest one such technique, and to use it to test the hypothesis that span recall is the product of a simple unitary process.Before proceeding to the experiment, it is perhaps important to consider the relevance for span recall of the findings and theories of supraspan research. It might be held that whether list length is just short This research was supported by National RelleUch Council of Canada Grant A8632 to Endel Tulving. Requests for reprints may be sent to Michael J. Watkins at the Department of Psychology, Princeton University, Princeton, New Jersey 08540. enough or just too long for perfect reproduction has little effect on the underlying mechanisms. We will argue that there are currently few if any good reasons for this view. A rather different view is that, unlike supraspan recall, the perfect reproduction of a list reflects a basic attribute of mental functio...