ABSTRACT Many classroom observation systems have been developed, including Flanders' interaction analysis. This system provides feedback to teachers about ten teacher-student verbal behavior categories. Data collected by observers are entered into a matrix which reveals ratios of the categories. Moskowitz's modification of interaction analysis (Foreign Language Interaction or Flint) expands Flanders to twenty-two categories. Both systems are deficient in accuracy, reliability, and practicality. Many be- havior categories are open to multiple interpretations. The number of categories observed, the rapid decisions made about frequently changing behaviors, and the necessity of recording data while verifying the passage of time make the observers' tasks difficult. The manner in which the data reliability is obtained is questionable. Researchers neglect instance by instance interactions; rather, by using Scott's coefficient, only summary data are considered. The time required to master these systems, the complexities of the category ratios and matrices, and the personal biases of teachers who observe themselves are all issues of practical concern. Further, there are serious methodological weaknesses in the research done thus far to evaluate these systems. Practitioners of interaction analysis and FLint must recognize the various problems of these methods before using them in research and training programs.