The AI literature contains many definitions of diagnostic reasoning (such as set-covering, abductive, consistency-based). However, diagnosis should not be seen as a problem with a unique definition. Instead, there exists a whole space of reasonable notions of diagnosis. These notions can be seen as mutual approximations. We use existing work approximate entailment to define notions of approximation in diagnosis. We show how such a notion of approximate diagnosis can be exploited in various diagnostic strategies. We illustrate these strategies by performing diagnosis in a small car domain example.