Discusses the number of observations necessary to obtain a trustworthy sample of classroom transactions by reanalyzing existing data. Eleven to twenty observations appear necessary for teachers' questions which require students to hypothesize. The reliability of a single observation was zero for four other questioning types. (CC)