Data were gathered from videotaped recordings of two preschool children engaged in unstructured free play over 12 days each. Observers coded behavior from the videotapes and completed a behavior rating scale for each child after every two observation sessions. Teachers also completed two behavior rating scales per child. Results indicated that at least three 30-min observation sessions were required to reliably represent a child's overall behavior. Moderate correlations were obtained when observations were compared with teachers' and observers' own ratings, indicating the behavior rating scale did an adequate job of reflecting actual observed behavior. The implications of these results for researchers and practitioners are discussed. (Contains 4 tables.)