Comorbidity between Attention Deficit Hyperactivity Disorder (ADHD) and reading disability (RD) is common; however, the heritability of this comorbidity is not well understood. This may be due to the complexity and heterogeneity of ADHD and RD phenotypes. Using alternative ADHD-RD sub-phenotypes instead of those arising from the DSM-IV may lead to greater success in the search for comorbid ADHD-RD susceptibility genes. Therefore, this study aims to refine ADHD-RD phenotypes into homogenous informative sub-phenotypes using latent class analysis (LCA). LCA was performed on 2,610 Australian twin families (6,535 individuals) in order to generate probabilistic genetically distinct classes that define ADHD-RD subtypes, including comorbidity, based on related symptom clusters. The LCA separated the phenotypes for ADHD and RD into nine classes. One class was unaffected; three classes demonstrated the three DSM-IV subtypes of ADHD, three subtypes showed different severities of RD, and two classes expressed a combination of RD and ADHD subtypes. LCA proved effective in refining the phenotypes of ADHD alone, RD alone, and ADHD-RD comorbidity, and its ability to classify them into homogenous groups based on clusters of symptoms, suggesting that the latent classes may be robust enough to use in molecular genetic studies.