The evolution of Aspect oriented (AO) software would degrade and modify its structure and its modularity. In this scenario, one of the main problems is to evaluate the modularity of the system, is the evolved AO software still has a good modularity or not? Unfortunately, this research area is not explored yet. This paper presents a history-based approach that detects modularity defects in evolved AO software. It is a two-step automated approach: 1) in the first step, it applies data mining over an AO software repository in order to detect logical couplings among its entities. It analyses fine-grained logical couplings between AO software entities as indicated by common changes. 2) These last are then analysed to detect modularity defects in the AO software system. The approach focuses on the evaluation of an AO system's modularity and points out potential enhancements to get a more stable one. We provide a prototype implementation to evaluate our approach in a case study, where modularity defects are detected in 22 releases of three well-known AspectJ systems: Contract4J, Health-Watcher and Mobile-Media. The results show that the approach is able to detect logical couplings among aspect entities, as well as modularity defects that are not easily (or not) detectable using static source code analysis.