Behavioral observation has become increasingly important in research in developmental and clinical psychology, as well as in program evaluation and instructional research. Several statistical approaches to assessing the dependability of behavioral observation were reviewed. The demand for accurate and efficient paradigms for behavioral observation has increased as psychologists view the method as a more viable alternative to traditional psychometric devices. Generalizability theory removes some ambiguity in interpretation by explicitly defining a universe of admissible observations and a universe of generalization. Analogies to traditional statistics were made using generalizability theory statistics to illustrate the similarity of concept and less ambiguous interpretation. Multivariate approaches were suggested for use with composite scores and situations where covariation between levels of a facet are of interest. Various approaches that have been suggested for behavioral observation were summarized. Continuous type scores were the primary focus of this study. More research is needed on agreement statistics when the nature of the task involved is classification on a nominal scale. (DWH)