This study investigated the complex influences of individual differences and school contexts on the growth of mathematical reasoning skills in high school, with particular emphasis on differences in academic preparation, demographics (i.e., race/ethnicity, gender and socioeconomic status), and school characteristics. Students' growth of mathematical reasoning skills, measured using standardized tests developed by the College Board (i.e., the PSAT/NMSQT and SAT), was examined by fitting a series of multilevel latent growth models to a sample of nearly 200,000 high school seniors from nearly 8,000 public schools. A number of hypotheses about patterns of change in the repeated measures of mathematical reasoning skills were examined at each of four levels of analysis: within individuals, between-individuals, across demographic groups and between schools. Results revealed that sophomore students, on average, started out at just over 48 points on a 20 to 80 mathematical reasoning scale After approximately two years of schooling they gained, on average, six and a halt score scale points. In the multiple group analyses of the growth models, mean differences in initial status and rate of growth were larger between White and Black students than between Male and Female students. The combined effects of family SES, general academic achievement, and mathematics background accounted for approximately 70 percent of between-student variability in initial status whereas only 15 percent of the variability in the rate of growth was accounted for by the student-level covariates. Multiple group analyses of this conditional model suggested that subgroup differences in mean initial status and rate of growth, although smaller than in the unconditional model, existed even after controlling for the effects of the student-level covariates. Finally, the multilevel school effects model suggested that individual differences in background characteristics made a difference at both the school and student levels; however, the effects were mostly on initial status rather than the rate of growth. The full multilevel latent growth model revealed school effects even after accounting for individual differences. Most importantly, school resources were not as influential as student body composition, highlighting the need to strengthen educational reform with social and economic development in high-need communities.