The benefits from practice follow a nonlinear function: Improvement is rapid at first but decreases as the practi-

The benefits from practice follow a nonlinear function: Improvement is rapid at first but decreases as the practitioner becomes more skilled (see, e.g., Thorndike, 1913). The idea that a simple nonlinear function might describe practice effects in a broad range of tasks was championed by Newell and Rosenbloom’s (1981) influential chapter entitled “Mechanisms of Skill Acquisition and the Law of Practice.” The “law of practice” in the title concerns the relationship between response time (RT) and number of practice trials. Newell and Rosenbloom examined data from a wide range of tasks. When they compared power and exponential functions as possible forms for the law of practice, power functions provided better fits than exponential functions in every case. The power function is now treated as the law of practice. In J. R. Anderson’s (1982) words, “one aspect of skill acquisition . . . distinguished . . . by its ubiquity . . . is the log-linear or power law for practice” (p. 397). A decade later, Logan (1992) echoed the same conviction: “The power law is ubiquitous. It occurs in virtually every speeded task” (p. 883). In accord with its status as a law, most research subsequent to the publication of Newell and Rosenbloom’s (1981) findings has assumed a power function, rather than testing to determine whether it provides a better description than other functions (e.g., Cohen, Dunbar, & McClelland, 1990; Kramer, Strayer, & Buckley, 1990; Logan, 1988, 1992), or has assumed that a power function holds for each component of performance (e.g., Delaney, Reder, Staszewski, & Ritter, 1998; Rickard, 1997). The power function’s status as a law has also made it a gold standard by which to judge the success of models of skilled performance, including ACT and related models (J. R. Anderson, 1982; J. R. Anderson & Schooler, 1991), the component power laws model (Rickard, 1997), network models (MacKay, 1982; Cohen et al., 1990), instance theories (Logan, 1988, 1992; Nosofsky & Palmeri, 1997), and Newell and Rosenbloom’s (1981) chunking model (see also Rosenbloom & Newell, 1987a, 1987b). Logan (1988) leaves no doubt about the importance of the form of the practice function for theories of skill acquisition: “The power-function speedup [is] a benchmark prediction that theories of skill acquisition must make to be serious contenders” (p. 495; see also Cohen et al., 1990, and Palmeri, 1997, for similar views). However, we contend that the evidence supporting a power law of practice is flawed. Although theories of skill acquisition model learning in individuals, the bulk of the evidence favoring the power law is based on fits to averaged data. There is little empirical evidence from individual learners for individual learning conditions that a power function describes skill acquisition better than does