This study modeled school and district effects in the mathematics scores of the Delaware Student Testing Program (DSTP) using hierarchical linear modeling. Three-level hierarchical models were fitted to estimate school and district effects in the DSTP mathematics scores and to examine the school and district variance with variables assessing student characteristics entered as predictors at each level. Data were collected in 4 waves, of which 3 were available for analyses: (1) 8,061 third graders from 1998 from 66 schools and 15 districts; (2) 8,066 fifth graders in 43 schools and 15 districts who took the test in 2000; and (3) matched data from 6,872 of the same students who took the third grade test in 1998 and the fifth grade test in 2000. Hierarchical linear modeling showed that the proportions of the variance at the school and district levels in the total variance of the DSTP scores for 1998 and 2000 were very small. Different patterns in the composition of variance at the school and district levels were observed for grade 3 and grade 5 students. For grade 3, the variance is predominantly at the school level, while for grade 5, the variance is nearly equal between the school and district levels. Results of the analysis also indicate that, in addition to the racial, gender, and socioeconomic status gaps commonly found in studies of mathematics performance, students who changed their schools, on average, had lower performance than those who stayed in the same school between grade 3 and grade 5. Results indicate the differential effect of Delaware schools on their students performance in mathematics on standardized tests, but the reasons for the differences are not known. (Contains 7 tables and 17 references.) (SLD)