One threat to the internal validity of coached-client analogue research provided a stimulus for this study of verification of experimental control of role performance. The study contrasts the judgments regarding one client's role performances which are made by both trained and untrained observers. The conclusion is that trained observers yield more accurate observations. The study includes an explanation of procedures for verifying role fidelity and performance consistency with a computerized system used by the trained observers. The author concludes with a discussion of implications for setting criterion levels for coached-clients, using observer feedback during actor training, and improving research replication. (Author)