This study examined the generalizability and dependability of observational estimates of middle and high school teachers' use of praise and reprimands. Frequency of behavior-specific praise, general praise, and total reprimands were collected across 67, 20-min observations that took place during class-wide instruction in general education classrooms. Generalizability theory was used to determine the number of observations needed to obtain dependable estimates of teacher behavior. Behavior-specific praise rates were consistently low. General praise rates were slightly higher and reprimand rates were notably higher and slightly more variable. Total reprimands had the strongest generalizability results and general praise had the weakest. Behavior-specific praise reached an acceptable level of dependability after 15 min, whereas general praise required a 35-min observation, and reprimand only required 5 min. Implications and future directions are discussed.