Using direct observation to assess physical activity permits the simultaneous collection of contextually-rich data in settings in which it occurs, and helps explain how physical activity is influenced by both physical and social environments. SOFIT (System for Observing Fitness Instruction Time) has been used to assess physical activity in physical education classes in over 2000 schools and 30 published papers. It uses time sampling to obtain simultaneous recordings of student activity levels, the lesson context in which they occurred (i.e., how lesson content was delivered, including time for fitness, skill drills, game play, knowledge, and management), and teacher interactions relative to promoting physical activity and fitness. Teacher gender, class gender composition, and lesson location, and number of students in class are also recorded. The activity codes (lying, sitting, standing, walking, and vigorous) have been calibrated using heart rate monitoring (Rowe et al., 1997) and validated using accelerometry (Scruggs et al., 2003). SOFIT has been used in at least 5 large National Institutes of Health studies. Baseline studies in over 1000 elementary schools indicate student activity engagement is about 37% of lesson time, compared to the 50% activity time recommended by Healthy People 2010. Results of three interventions studies using an active curriculum, staff development, and on-site consultations show that activity engagement in classes can be increased by about 18% without increasing the frequency or duration of PE lessons. Reliabilities for observations by trained independent observers typically exceed 90% agreement on all SOFIT categories. This presentation shares information on procedures used to train and recalibrate SOFIT observers and provides practical examples of how data collected using SOFIT can be used in formative, process, and outcome evaluations. Keyword(s): exercise/fitness, measurement/evaluation, physical activity