Procedures are presented for training classroom observers to collect data in naturalistic research projects. Training methods are directed toward the collection of highly reliable observational data. Pre-classroom training prepares observers for data collection by working with the coding manual, video-tapes, and written dialogues, and through group discussions with the trainer. A list of practical guidelines for coders to follow while collecting data helps to minimize observer effects, maintain good relations with school personnel, and insure high quality data. Techniques for training in the classroom are described, in addition to ways of dealing with difficulties that may arise in using the observational system. A discussion of inter-coder reliability describes methods of obtaining reliability estimates and also analyzes the problems that may arise when attempting to establish a high degree of reliability among coders. The training procedures presented in this report can be used in either small- or large-scale research projects and are appropriate for training observers to use both low- and high-inference observational systems. (Author)