MDE and software evolution call for model-level design support, that includes reasoning capabilities such as query answering, verification and validation, static analysis and model transformation. Automation of all activities requires wellde fined semantics for models. This is particularly important for the class diagram model, which is central in UML. However, since UML specification is verbal and imprecise, the exact meaning of many class diagram constructs and their interaction is still obscure. There are major problems with the inter-association constraints subsets, union, redefinition, as- sociation specialization, association-class specialization and XOR. Although their standard semantics is ambiguous and their interaction unclear, the UML meta-model intensively uses these constraints. Moreover, some of these interactions have been declared in the UML meta-model as variation points. The paper discusses possible semantics of inter-association constraints (subsets, union, redefinition and association-class specialization), analyzes their interaction, and suggests coherent semantics that minimizes contradictions with the metamodel (association specialization and XOR are left out, due to space limitations). The paper also introduces rules that enforce model correctness. This paper is the first to provide an inclusive analysis of all inter-association constraints.