Divergence and non-determinism play a fundamental role in the theory of computation, and their combined effect on computational equality deserves further study. By looking at the issue from the point of view of both computation and interaction, we are led to a canonical equality for non-deterministic computation, revealing its rich algebraic structure. We study this structure in three ways. First, we construct a complete equational system for finite-state non-deterministic computation. The challenge with such a system is to find an equational alternative to fixpoint inductionà laMilner. We establish a negative result in the form of the non-existence of a finite equational system for the canonical equality of non-deterministic computation to support our approach. We then investigate infinite-state non-deterministic computation in the light of definability and show that every recursively enumerable set is generated by an unobservable process. Finally, we prove that, as far as computation is concerned, the effect produced jointly by divergence and non-determinism is model independent for a large class of process models.We use C-graphs, which are interesting in their own right, as abstract representations of the computational objects throughout the paper.