Error patterns have been found to be sensitive to cognitive status, but the relationship between aging and error patterns remains unclear, and may differ as a function of gender, education, and whether a task is verbal or nonverbal. The present study examined the error patterns of normal elderly individuals on a verbal measure of set-shifting and rule induction to determine whether demographic variables, that is, age, gender, and education, influenced test performance. The sample of 109 individuals, 38 males and 71 females, ranging in age from 54 to 89 years with 6 to 19 years of education, was assessed on the Classification subtest of the Test of Verbal Conceptualization and Fluency, a verbal measure of set-shifting and rule induction. Subjects' protocols were scored for perseverative, nonperseverative, and random errors, tabulated, and analyzed. Multivariate analysis of covariance with education as the covariate as well as other statistical tests revealed nonsignificant relationships between error scores and age, gender, and education. Years of education, however, showed a significant correlation with a reduction in random responses. Results are interpreted based on Horn's (1978) fluid-crystallized explanations of changes in intelligence with advancing age.