This paper advocates the use of exploratory factor analysis (EFA) for Q-matrix specification in cognitive diagnostic assessment (CDA) of language comprehension. We propose that integrating EFA results into Q-matrix specification improves the psychometric quality of CDA classification results with reference to Q-matrices based solely on human judgment. To test this hypothesis, we conducted EFA on the response data of 3,607 test takers in a large-scale EFL reading test. We then specified two Q-matrices, one combining EFA results with subject matter expert judgment (SME+EFA Q-matrix) and the other based solely on expert judgment (SME Q-matrix). We fit the G-DINA models to the response data using the two Q-matrices. The results showed that the G-DINA model based on the SME+EFA Q-matrix fit the data better and yielded clearer attribute distinction and higher classification accuracy. To examine the generalizability of the results, the same models were fitted to a second dataset with 1,577 test takers. The SME+EFA Q-matrix again yielded better results than the SME Q-matrix, but heterogeneity between the samples resulted in inferior classification accuracy relative to the first dataset. The findings made a strong case for using EFA to complement expert judgement in Q-matrix specification for CDA studies in language assessment.