This nonexperimental, longitudinal field study examines the extent to which variation in observed classroom supports (quality of emotional and instructional interactions and amount of exposure to literacy and math activities) predicts trajectories of achievement in reading and math from 54 months to fifth grade. Growth mixture modeling detected two latent classes of readers: fast readers whose skills developed rapidly and leveled off, and a typical group for which reading growth was somewhat less rapid. Only one latent class was identified for math achievement. For reading, there were small positive associations between observed emotional quality of teacher-child interactions and growth. Growth in math achievement showed small positive relations with observed emotional interactions and exposure to math activities. There was a significant interaction between quality and quantity of instruction for reading such that at higher levels of emotional quality there was less of a negative association between amount of literacy exposure and reading growth. (Contains 5 tables, 4 figures, and 1 note.)