Despite the increasing interest in and need for test information for use in instructional practice and student learning, there have been few attempts to systematically link a diagnostic approach to English for academic purposes (EAP) writing instruction and assessment. In response to this need for research, this study examined the extent to which the diagnostic information generated by the Reduced Reparameterized Unified Model (Reduced RUM; Hartz, Roussos, & Stout, 2002) was a discriminant, accurate, and reliable method of determining student performance in EAP writing. Ten English as a second language (ESL) teachers assessed 480 Test of English as a Foreign Language[TM] Internet-based test (TOEFL[R] iBT ) independent essays using the Empirically-derived Descriptor-based Diagnostic (EDD) checklist, which consisted of 35 concrete, fine-grained descriptors. The resultant ratings were then analyzed using "Arpeggio," the estimation software of the Reduced RUM. The findings showed that the skills diagnosis approach not only classified skill masters and non-masters accurately and reliably, but that it also had high discriminant function, with only a small number of students classified into flat profiles. However, concerns were raised with regard to the diagnostic power of some descriptors and the stability of one writing skill parameter (i.e., mechanics). The theoretical and pedagogical implications of the use of diagnostic assessment in EAP writing are discussed in this paper, as are suggestions for future research in applying psychometric diagnostic assessment models to performance assessment. (Contains 3 notes, 9 tables, and 10 figures.)