One practical approach to using ontologies for knowledge representation and reasoning is the approximation through typed feature structures. The non-deterministic multiple inheritance hierarchies used for ontologies can be mapped immediately to typed feature structures. An important inference operation is default unification which computes the maximally compatible combination of new information (cover ) and old information (background). It is fundamental to ensure that such an operation does not limit the design of the ontology. In this work we provide an extension to default unification. Firstly the operation produces feasible results in multiple inheritance hierarchies that are nondeterministic. Secondly the specification of the algorithm guarantees that well-formedness is not violated.