On the basis of the ACT* production system theory of skill acquisition (Anderson, 1983a(Anderson, , 1987, I generated predictions concerning the role of two proposed classes of working-memory limitations in procedural learning. Individual differences analyses of a laboratory procedural learning task tested these predictions. As hypothesized, measures of controlled attention in working memory predicted initial declarative rule acquisition and proceduralization, and measures of automatic activation in working memory predicted indexes of later production composition and strengthening. Results supported a distinction between two working-memory capacity constructs that impose unique limits on processes of skill acquisition.Production system models provide a useful theoretical tool for understanding the acquisition of procedural skills. Anderson (1982Anderson ( , 1983aAnderson ( , 1987 described the acquisition and improvement of complex procedural skills such as geometry problem solving in terms of the ACT* production system theory. Likewise, Kieras and Bovair (1986) found that a similar production system facilitated the understanding of procedural learning about a physical device.According to the production system framework described by Anderson (1983aAnderson ( , 1987, procedural skills are initially developed through the interpretation of declarative knowledge, much like one would interpret and execute steps of an unfamiliar recipe. Executing procedures at this declarative stage is typically effortful, slow, and error-prone. With continued interpretation of declarative knowledge as procedural actions, however, another memory representation is thought to develop in the form of productions. Unlike the slow interpretation of declarative procedures, a production's action is immediately executed whenever necessary conditions are met. This establishment of direct condition-action associations, defined as proceduralization by Neves and Anderson (1981), is used to explain performance improvements during initial practice of a procedural skill.I wish to sincerely thank Raymond Christal, William Tirre, Valerie Shute, Scott Chaiken, and especially Patrick Kyllonen for suggestions during the conduct of this research and for comments on earlier drafts of this article. I also acknowledge the many helpful suggestions by two anonymous reviewers. Finally, I am indebted to the many programmers from OAO Corporation who worked on different aspects of this research.

The construct of working memory (WM), as described by Baddeley and Hitch (1974), has had a far-reaching impact on theories of cognition. For years, Baddeley's model was the predominant theory of WM, and it remained relatively unchanged, with its two storage components for visuospatial and phonological information and a central executive component for coordinating storage and various processing functions. 1 A defining feature of all three of these subcomponents has been their attention-driven nature. In fact, at one point, Baddeley (1993) reflected that it may have been more appropriate to use the term working attention, rather than working memory. The term attention is used by Baddeley, as well as by many other authors within the WM literature, to refer broadly to effortful processes that range from rehearsal procedures to executive functions, such as planning, organizing, and controlling cognitive actions. In this article, we rely on the broad distinction between deliberate, attention-driven processes of WM that operate within the awareness of the individual and processes that operate outside of conscious awareness, such as implicit memory processes that underlie various forms of priming.Various theorists have noted that relatively smallcapacity, attention-driven storage components of WM fail to explain complex cognitive activities such as language comprehension, in which a great deal of information needs to be temporarily available for processing (e.g., Anderson, 1983;Cowan, 1999;Ericsson & Kintsch, 1995;Just & Carpenter, 1992;Kintsch, Patel, & Ericsson, 1999). Corresponding to this criticism, alternative conceptualizations of WM have been proposed that allow for more information to be available for processing at any given moment. Most define this greater availability in terms of long-term memory (LTM) activation, or the temporary increase in retrieval strength of existing memory representations. These models differ in important respects, but of interest here is their common inclusion of long-term memory structures that are temporarily available for processing but that are not in the current focus of attention. Most of these alternative models postulate that both attentiondriven WM processes and automatic LTM activation or retrieval processes effectively define capacity limits that constrain complex processing activities. Furthermore, some suggest that the LTM activation processes are capacity limited in a different manner than the attention-driven processes. However, direct empirical evidence regarding these issues is limited. LTM Processes in WM ModelsA variety of WM models that represent alternatives to Baddeley's model have been proposed within different theoretical contexts (see Miyake & Shah, 1999). We will describe only a few of these that are most prominent in their inclusion of active-but-unattended LTM elements in WM. Virtually all WM theorists acknowledge the contribution of LTM knowledge structures to WM performance, but some have proposed models in which LTM processes are central to the definition...

Relations among aptitude, spatial task solution strategy, and task performance were explored through mathematical models of performance latency. Single-strategy and strategy-shifting models were tested individually for a stratified random sample of 30 male high school and college subjects. Different models fit different subjects on each of three task steps (encoding, synthesis, and comparison), suggesting that different subjects used different strategies for solving the same items. Some of the best fitting models specified that subjects frequently and flexibly switched strategies in keeping with variations in item demands. This was considered a form of adaptive, within-task learning. For the encoding and synthesis steps, significant performance differences were found among subjects using different strategies. Inspection of aptitude profiles suggested that aptitude may have affected strategy choice. The importance of strategy-shift models for understanding complex problemsolving processes and for representing adaptation and flexibility in intelligent performance is discussed. Do individuals solve cognitive tasks in different ways? Or are individual differences primarily a function of how effectively subjects execute a common solution strategy?Current research on aptitude uses information processing models to identify components of cognitive performance. Such models have suggested that individuals dif-

No abstract