Classroom researchers are frequently urged to provide evidence of the reliability of their data. In the case of observational data, three approaches to this have emerged: observer agreement, generalizability theory, and measurement error. Generalizability theory provides the most powerful approach given an adequate data collection design, but observational data provide an opportunity to obtain direct measures of the magnitude of measurement error in a way that is not possible with other kinds of data used in educational research. Two methods of assessing measurement error are demonstrated--by direct measurement and by regression estimates based on mean score. An example, using 105 primary children in self-contained classrooms, investigates the influence of classroom processes on the reading performance of learning disabled students. The argument is made that there is no one correct way of expressing the adequacy of data for all purposes; instead one must turn attention to the specific indicators of data quality that are most relevant to the intended use of the data. A table outlines indicators of adequacy of data. (Author/JAZ)