In this study, generalizability theory was used to examine the extent to which (a) time-sampling methodology, (b) number of simultaneous behavior targets, and (c) individual raters influenced variance in ratings of academic engagement for an elementary-aged student. Ten graduate-student raters, with an average of 7.20 hr of previous training in systematic direct observation and 58.20 hr of previous direct observation experience, scored 6 videos of student behavior using 12 different time-sampling protocols. Five videos were submitted for analysis, and results for observations using momentary time-sampling and whole-interval recording suggested that the majority of variance was attributable to the rating occasion, although results for partial-interval recording generally demonstrated large residual components comparable with those seen in prior research. Dependability coefficients were above .80 when averaging across 1 to 2 raters using momentary time-sampling, and 2 to 3 raters using whole-interval recording. Ratings derived from partial-interval recording needed to be averaged over 3 to 7 raters to demonstrate dependability coefficients above 0.80.