An issue facing researchers who study very select populations is how to obtain reliability estimates on instruments. When the populations and resulting samples are very small and select, the ability to obtain reliability estimates becomes very difficult. As a result, many researchers ignore reliability concerns and forge ahead with data collection. In response to this concern, concepts associated with the model of social cognition of A. Bandura and the model of systematic desensitization of J. Wolpe were applied to 90 undergraduates who completed the 10-item, Likert-type Communication Satisfaction Scale (CSS). The CSS is designed to assess the attitudes of intubated patients in a hospital intensive care unit. The purpose of this study was to explore the potential use of imagery and social cognition to estimate instrument reliability for very selective samples. Stimuli (text, auditory, and visual) were provided to enable subjects to imagine being an intubated patient. Internal reliability calculations (Cronbach's alpha index) revealed an estimate of 0.83 for the entire scale. The results are discussed within a social cognition and a measurement framework. These results may have profound implications for estimating instrument reliability that cannot be tested feasibly in large samples prior to implementation. While the resulting reliabilities cannot be directly applied to the intubated sample, the procedure may provide critical feedback to researchers and instrument developers prior to the actual administration of the instrument. The CSS is included. (Author/SLD)