A laboratory analogue of naturalistic observation was used to examine the relationship of observer drift to instructional set and experimenter status. Results indicated a highly significant decrease in observer reliability coinciding with the shift from training to data collection. (ST)