Identifying sources of variation has been used extensively in educational research as a tool to identify potential drives of variances in student achievement. However, prior research predominantly relied on findings from national- or international-level data, and thus their conclusions remain very broad-based. This study contributes new insight by assessing if and where there is variation in standardized testing performance for entire populations of cohorts of students in a single, large urban school district in the United States. Specifically, this study evaluates variance in Stanford Achievement Test Ninth Edition (SAT9) reading and math scores for all elementary school students in the School District of Philadelphia over four academic years and within three analytical levels of the educational experience—student, classroom, and school. To do so, this study employs three-level hierarchical linear modeling (HLM) to determine how the overall variance in testing performance can be partitioned within classrooms, between classrooms, and between schools. The initial results indicate that the overwhelmingly largest contributor to total variance in achievement is within classrooms at the student level. However, incorporating a full span of covariates into a three-tiered model of student achievement explains the majority of the between classroom and between school variance, though only half of the within classroom variance. Implications are discussed.