This study examined whether Matthew effects were evident in developmental patterns of reading and mathematics skills from middle childhood to adolescence. We obtained standardized reading and mathematics scores at Grades 3, 5, 7 and 9 for full cohorts of students in two Australian states, NSW (N = 88,958, 48% female) and Victoria (N = 65,984, 49% female). Latent growth curve models were used to identify the best-fitting longitudinal trajectories of reading and mathematics, and to examine whether cumulative (i.e. a Matthew effect), compensatory, or stable interindividual differences characterized development in each domain. For both reading and mathematics, and in both samples, growth decelerated as grade levels increased, with latent basis models fitting the data better than linear models. Negative intercept-slope covariances, and decreasing variances at increasing grades in both domains indicated compensatory growth patterns, rather than cumulative patterns or Matthew effects. These results indicate that students with below average achievement at Grade 3 make greater gains to Grade 9 than their initially higher-achieving peers. Modeling growth trajectories in two longitudinal population datasets allows strong tests of theorized growth patterns for both reading and mathematics, and presents insights about developmental change in academic skills from middle childhood to adolescence.