The present study evaluated the effects of stimulus complexity and rule usage on a visual perspective-taking task. Preschoolers, first, third, and fifth graders, and adults were shown arrays of dolls and performed a series of perspective-taking tasks. Errors decreased with age, and more errors occurred with the more complex visual arrays. A significant number of errors were made in self-view trials, especially by the preschoolers, showing that the ability to relate an array to a pictorial representation of it is not perfect. A conditional probability analysis showed that most egocentric errors were not due to an inability to relate the array to pictorial representations, but rather to a lack of mastery of Flavell's different positions--different views rule. When the array was covered, however, even first graders showed almost perfect mastery of this rule. There were also task effects on the use of Flavell's same position--same view rule: children performed better for a task involving self and other than for 2 others. Response latencies and effects for the observer's relative position provided evidence for a new rule: opposite positions--opposite views. In addition, front and back views of the dolls were significantly easier than the side views, which suggests a role of labeling or stimulus-discrimination skills.