We explored the usefulness of first and second grade reading measures and responsiveness criteria collected within a response-to-intervention (RtI) framework for predicting reading disability (RD) in third grade. We used existing data from 387 linguistically diverse students who had participated in a longitudinal RtI study. Model-based predictors of RD were analyzed using logistic regression; isolated measure/criteria combinations for predicting RD were analyzed using classification analysis. Models yielded superior classification rates compared to single measure approaches and did not systematically misclassify English learners. However, particular first and second grade measure/criteria combinations also showed promise as isolated predictors of RD in word reading/text fluency. Model-based approaches were required for acceptable classification of students with RD in comprehension. Although the former finding is promising for early identification of students in need of more intensive instruction in lexical or fluency-based skills, the latter finding reaffirms literature attesting to the complexity of RD in comprehension and difficulty of predicting deficits using early measures of reading, which primarily assess word reading skill. Results replicated well with an independent sample, thus enhancing confidence in study conclusions. Implications regarding the use of RtI for predicting RD are discussed.