L. L. Jacoby, J. P. Toth, and A. P. Yonelinas (1993) advocated a process-dissociation procedure for estimating the contributions to task performance of consciously controlled (R) versus automatic (A) memory processes. The procedure relies on the strong assumption that memory-guided performance attributable to R is stochastically independent of that attributable to A. Violations of this independence assumption can produce artifactual dissociations between estimates of R and A. Such artifactual dissociations were obtained in a series of word-stem completion experiments: R increased with presentation duration, whereas A, paradoxically, decreased. Direct evidence for nonindependence was obtained from correlations between R and A in each of the experiments. These results suggest that the independence assumption was violated, and other applications of process dissociation should not be taken at face value without a thorough evaluation of independence.