The purpose of this study was to compare how learning outcomes are influenced by a number of key student composition (language background, socioeconomic status, ethnicity, and gender) and school context variables on a direct writing performance assessment. The sample was randomly selected from a population of 13,604 third graders in 175 public elementary schools in Hawaii. Students completed the Stanford Achievement Test Edition 8 (indirect writing) and the Stanford Writing Assessment Edition 1 (direct writing assessment). Students included in the sample of 3,300 had completed both the direct and indirect writing assessments. The multilevel model included student-level controls and school-level controls and the student-level variables of ethnicity, language background, socioeconomic status, and gender and the school-level variables of percent of students with limited English skills and percent of low socioeconomic status students. Results show that the set of equality variables accounted for a relatively small proportion of variance at both the student and school levels on the direct writing assessment, with only slightly more than 40% of the variance in direct writing scores attributed to student composition and school context factors. Findings suggest that direct writing assessment appeared to have measured a more diverse set of skills and to yield scores that were less affected by variables that were outside the school's control. Implications for performance assessment are discussed. (Contains 2 tables and 60 references.) (SLD)