It is suggested that the low reliability of five low-inference and two high-inference classroom observation systems are due not to lack of agreement among rates, but to apparent variation in teacher behavior across occasions. (Author/MJB)