Classroom observation has the potential of obtaining valuable information regarding teacher and child behavior. This study examines the accuracy of observers coding a standard stimuli. In this procedure the observer's bias is examined, as well as the confidence that can be placed in the observation code itself. Through these procedures the exact nature of the confusion of codes can be identified. In order to avoid the problems encountered with the paired observer method, an attempt was made to assess the accuracy of observers through the use of controlled videotape examples which allow each interaction (a frame) and sequences of frames to be analyzed for accuracy. Ten videotaped skits were produced to present concise, clear examples of each code used in recording classroom interactions on an observation instrument. Confusability (low observer agreement) matrices were constructed by tallying the observer code sequences. Results of the confusability study identify the specific codes that appear to be reliable as well as those that are confused and need to be redefined. Inter-rater accuracy and videotape simulation accuracy are compared. While the two systems of examining observer accuracy do yield some different information, it is not contradictory, and the videotape system is easier to interpret. (Author/RC)