The influence of day-to-day variation in teacher performance on the reliability and validity of teacher assessment was examined. An attempt was made to identify and quantify sources of score variation attributable to differences in teacher performance, day of observation, observers, and test subscales; and to determine their effects on reliability and validity of decisions made. Data were collected from a field test of the revised Teacher Performance Assessment Instruments (TPAI). Thirty-nine seventh grade science teachers in a large Georgia school district were observed twice: (1) by two observers at the same time on the same day; and (2) by two observers on different days. Each teacher taught a two-week unit on scientific problem solving; student achievement was measured with pretests and (the Group Assessment of Logical Thinking) and posttests (Middle Grades Integrated Process Skills Test. The estimated expected posttest score was substracted from the actual score to provide an index of teacher effectiveness. A four-facet fully-crossed design was used, including teachers, observers, day of observation, and performance indicators as sources of variation. Results suggested that additional observation time enhanced validity, and that daily variation was a greater source of error than observer differences. (GDC)