Ever since Ebbinghaus (1885Ebbinghaus ( /1913, students of memory have sought to understand human sequence learning. Whereas early work analyzed the acquisition of long lists of items through repeated study and test trials (see Harcum, 1975, for a review), more recent studies have favored the analysis of immediate recall of relatively short lists (Brown, Preece, & Hulme, 2000;Burgess & Hitch, 1999;Cowan, Saults, Elliott, & Moreno, 2002;Farrell & Lewandowsky, 2002;Henson, 1998;Knoedler, Hellwig, & Neath, 1999;Li, Schweickert, & Gandour, 2000). This shift in attention reflects a desire to move to a finer grain of analyses-from measures of overall recall and learning to the analysis of the distribution of errors across individual list positions. This article bridges these two approaches by bringing a more detailed approach to the analysis of serial learning. Through analyses of both item and order gains and losses over trials, we show that one can distinguish among hypotheses of serial learning that are indistinguishable with analyses of mean recall accuracy alone.The effect of repetition on learning is most often measured by plotting a learning curve. Consider the free recall task, in which participants are asked to recall a justpresented list in any order. In this task, recall probability increases logarithmically with the number of study trials (Tulving, 1962). The logarithmic form of the learning curve could arise from different underlying processes. For example, participants might maintain the items they have already recalled, but pick up fewer and fewer new items over trials. Alternatively, participants might pick up new items at a constant rate, while forgetting a percentage of the items that they had previously recalled. This would lead to a higher absolute number of items being forgotten over trials, such that the net gain of recalled items still increases over trials. Looking solely at the learning curve, one cannot distinguish between these possibilities. But by examining what happens to individual items from trial to trial, one can test these hypotheses.In a classic analysis of the learning curve in free recall, Tulving (1964) examined these trial-to-trial transitions of individual items. He reminded us of two important features of episodic list-learning experiments. First, in most studies, which use words as stimuli, participants are not learning items per se; rather, they are learning that these items, mini-events in the experiment, occurred in a particular spatiotemporal context defining the study list. Tulving (1964) also noted that if queried immediately following an item's appearance on the list, participants would not have difficulty recalling that item. Thus, the process of learning is inextricably tied to the process of forgetting (and vice versa;Krueger, 1929). Tulving (1964) decomposed the learning curve into four mutually exclusive types of information. Consider the fate of a study item on a given learning trial. In one case, an item that was not recalled on the previous trial is recalled on...