This article reports on findings based on analyses of a unique dataset collected by ACT that includes information on student achievement in a variety of subjects at the high-school level. The authors examine the relationship between teacher effect estimates derived from value-added model (VAM) specifications employing different student learning assumptions. Specifically, they compare teacher effectiveness estimates derived from a traditional lagged score VAM using pretests and posttests in a single subject area to those derived from VAM specifications employing a cross-subject student fixed-effects approach. The latter approach offers advantages for teacher evaluation systems due to significantly reduced data requirements; however, there is evidence that both the estimated effect size of teacher quality as well as estimates of individual teacher performance vary depending on the VAM methodology. In particular, teacher effects identified based on within-student cross-subject variation results in significantly smaller effect size estimates than do those generated from a more traditional lagged score model. The correlation across model specification ranges from 0.25 to 0.96 depending on the subject area. (Contains 35 notes and 5 tables.)