A laboratory analogue of naturalistic observation was used to examine the relationship of observer drift to instructional set and experimenter status. 3 instructional sets (no check, random check, and spot check) and 2 levels of experimenter status were studied. Results indicated a highly significant decrease in observer reliability coinciding with the shift from training to data collection This performance decrement was observed in all 3 instructional set conditions. Within the spot-check condition, reliability on spot-check days was found to be significantly greater than mean reliability immediately before and after spot checks. Further results revealed that observers trained by the high-status experimenter performed less reliably than observers trained by the other 2 experimenters. The possible implications of these results for future observational research and suggestions for minimizing observer drift are discussed.