College faculty are unlike other teachers in that they are not trained to teach and have little supervision. Effectiveness of language instruction is more difficult than most to evaluate. Contemporary models of supervision tend to emphasize training, not evaluation, even for teaching assistants. Some see teaching as an art, not susceptible to empirical study. Much current second language teaching research is insubstantial. Existing approaches to classroom observation include rating scales, checklists, time-based or event-based tabulations, and ethnographic methods. Some methods are high-inference and some low-inference, but results most often involve a high degree of inference. More attention should be given to the observation/evaluation instruments used. Minimum requirements for a useful instrument include practicality, reliability, and validity. The profession should develop profession-wide rather than institution-based instruments reflecting relevant theory and practice. There exist models for development of such a measure, and it is possible to make one that is dynamic, capable of adaptation to new research evidence. Issues that might be addressed in creation of an empirically-based teaching evaluation instrument include quantity and quality of teacher talk, comprehensible input, formal grammar practice, error correction, teacher-student interaction, and teacher experience. Contains 36 references. (MSE)