We conducted generalizability studies to examine the extent to which ratings of language arts performance assignments, administered in a large, diverse, urban district to students in second through ninth grades, result in reliable and precise estimates of true student performance. The results highlight three important points when considering the use of performance assessments in large-scale settings: (a) Rater training may significantly impact reliability; (b) simple rater agreement indices do not provide enough information to assess the reliability of inferences about true student achievement; and (c) assessments adequate for relative judgments of student performance do not necessarily provide sufficient precision for absolute criterion-referenced decisions. (Contains 4 figures, 6 tables and 1 footnote.)