Twenty-eight mature women were recruited from the community and trained in a 21 category observation code of family interaction. Observers were assigned randomly to three experimental groups and given different expectancy rationales about the outcomes of the studies for which they would be collecting data. All groups were told they would be observing a family under a father-present and father-absent condition. One group was led to expect an increase, another a decrease, and a third no change in the rate of deviant behavior for the boys in the family as conditions changed from father-present to father-absent. None of the groups were told they would be observing identical videotape recordings of family interaction permitting comparison of observation data across groups. Results indicated that the expectations of experimental outcomes differed significantly for the three groups. However, observers were totally unbiased in their reports of deviant behavior in group comparisons. Failure to obtain evidence for observer bias in spite of the demonstrated manipulation of observer expectations was attributed to the precautions taken to assure high levels of observer accuracy throughout the study. (Author)