Recently, Smith, Thelen, and colleagues proposed a dynamic systems account of the Piagetian "A-not-B" error in which infants' errors result from general processes that make goal-directed actions to remembered locations. Based on this account, the A-not-B error should be a general phenomenon, observable in different tasks and at different points in development. Smith, Thelen, et al.'s proposal was tested using an A-not-B version of a sandbox task. During three training trials and three "A" trials, 2-year-olds watched as a toy was buried in a sandbox at Location A. Following a 10-s delay, children searched for the object. Across five experiments, children's (total N = 92) performance on the A trials was accurate. After the A trials, children watched as a toy was hidden at Location B, 8 to 10 inches from Location A. In all experiments, children's searches after a 10-s delay were significantly biased in the direction of Location A. Furthermore, this bias toward Location A decreased with repeated trials to Location B, as well as when children completed fewer trials to Location A. Together, these data suggest that A-not-B-type errors are pervasive across tasks and development.