Describing the variation in test scores between and within school districts is critical for: (1) for policy-related and descriptive work that investigates the sorting of students among districts and the differential effectiveness of those districts; and (2) for methodological work planning future experiments or interventions. Intraclass Correlations (ICCs) and Coefficients of Variation (CVs) are two complementary ways to describe test score variation. ICCs describe the proportion of variance in test scores that is between (rather than within) school districts or schools. CVs describe the extent of heteroscedasticity in district (or school) test score distributions. The most straightforward method of calculating ICCs and CVs is to use student-level data to directly estimate the means and variances of district or school test score distributions. In this study the authors investigate three interrelated research questions: (1) how much between-district variation exists across U.S. states?; (2) what are they key patterns across grades, subject, and years (within or across states) in the between-district variation?; and (3) what state-level factors are associated with a state having more-or-less observed between-district variation? This study leverages the use of ordered probit models to recover distributional information from coarsened test score data proposed by Reardon et al. This method was applied to a large data set provided by the National Center of Education Statistics (NCES) through a restricted data license. Under the No Child Left Behind (NCLB) legislation, states are required to report aggregated test score results to the U.S. Department of Education, through a system called EdFacts. The authors found substantial variation in ICCs and CVs across states. One table and 2 figures are appended.