Currently in applying IMS LD, a unit of learning has to be completely defined and then it will be executed exactly following the definition. However, many actual learning processes involve uncertainties and dynamical changes. Due to imprecise definition and unexpected run-time events, probably a learning process cannot be executed as planned in the learning design. That is, exceptions may raise and the execution of the normal process model cannot progress. So far the issue of exception handling hasn?t been seriously addressed in IMS LD community. In this paper the technical requirements for detecting and handling exception are analyzed. Then, the paper presents a proposal enabling learning designers to define exception handling rules by using the same formalisms for modeling normal learning processes.