During an extensive inservice program, teachers and aides in 20 kindergarten and first grade Follow Through Classrooms were observed for mastery of the Direct Instruction Model. Two trained observers used the Direct Instruction Supervisor Code to record behavior in the fall, winter, and spring. Subjects were observed while teaching a 12 minute lesson in reading or oral language. Mastery was hierarchical: most subjects mastered the two basic components (formats and signals) within two months, and the remaining components (corrections, pacing, and student accuracy) by May. Reliability and validity of the Code, and students' achievement gains were also examined. Code scores had high interobserver reliability and low to moderate reliability between fall and winter observations. Scores had moderate concurrent validity with supervisor ratings and discriminated well between high and low implementers. All aides and teachers scored poorly on Loucks' Levels of Use interview for assessing educational innovations: its validity for a performance based model, such as Direct Instruction, was questioned. Correlations with scores on the Comprehensive Tests of Basic Skills were low for pacing and formats, and moderate for the remaining components. Tracing skill mastery over long periods (1-2 years) can offer valuable information for inservice training. (CP)