This study compared the performance of 946 8th-grade students with different language proficiencies (limited English proficient [LEP], fluent English proficient [FEP], and initially fluent in English [IFE]) and language backgrounds on a 35-item math test (from the 1996 National Assessment of Educational Progress (NAEP) Grade 8 Bilingual Mathematics booklet) when different test accommodations were provided (original items, linguistically Modified English items, original items plus Glossary, original items plus Extra Time, original items plus Glossary and Extra Time). A reading test (11 items from the NAEP 1992 Grade 8 Reading assessment) and a language background questionnaire (Abedi, Lord, and Plummer, 1995) were also administered. For the entire sample, providing extra time for the math test resulted in a 1-point increase in student mean scores (14.68 for original items, and 15.64 with extra time). When a glossary and extra time were provided, the mean score was more than 2 points higher (mean 17.08). For the entire sample, no significant difference was found when items were linguistically modified (mean 14.23) or a glossary was provided without extra time (mean 14.53). Major findings include the following. Students designated LEP by their schools scored, on average, more than 5 points lower than non-LEP students on the math test. In comparison with scores on the original NAEP items, the greatest score improvements, by both LEP and non-LEP students, were on the accommodation version that included the Glossary plus Extra Time. LEP students scored higher with all types of accommodation except Glossary only. Most accommodations helped both LEP and non-LEP students; however, the only accommodation that narrowed the score difference between LEP and non-LEP students was Modified English. Students who were better readers, as measured by reading test scores, achieved higher math scores. These differences and relative impacts need to be considered and investigated further before accommodation strategies are adopted for large-scale assessments. (Contains 90 references.) (MM)