Due to the implicit characteristics of learning disabilities (LDs), the identification or diagnosis of students with LDs has long been a difficult issue. In this study, we apply rough set theory (RST), which may produce meaningful explanations or rules, to the LD identification application. We also propose to mix samples collected from sources with different LD diagnosis procedure and criteria. By pre-processing these mixed samples with some simple and readily available clustering algorithms, we are able to improve the quality of rules generated by RST. Our experiments also indicate that RST performs better in term of prediction certainty than other rule-based algorithms such as decision tree and ripper algorithms. Overall, we believe that RST may have the potential in playing an essential role in the field of LD diagnosis.