The use of the intraclass correlation in determining reliability is discussed and shown to be both appropriate and simple to use in the case of an observational measure, provided that observations are made on at least two occasions. The interpretation of such coefficients is explained in terms of generalizability theory, and real data are used to demonstrate how such coefficients can be interpreted and computed. Finally, an empirical study is described which investigates the effect on reliability of varying the number and length of the observation periods. (Author)