Observational data were collected over a period of 120 school days from 11 elementary level, self-contained, learning disabilities classrooms (N = 105), using the Student-Level Observation of Beginning Reading (SOBR) system. Various studies were conducted to evaluate the generalizability of 10 observational measures. The major findings were that: observers were well-trained such that reliable and unbiased observations of student behavior were obtained through the use of a single observer; student behavior as observed over 30 occasions was extremely stable; and a high stability coefficient would be obtained even for five one-hour observation sessions randomly sampled from the school year. Implications for future generalizability studies of observational data are also discussed.