Abstract. We show that non-determinism simplifies coding certain problems into programs. We define a non-confluent, but well-behaved class of rewrite systems for supporting non-deterministic computations in functional logic programming. We show the benefits of using this class on a few examples. We define a narrowing strategy for this class of systems and prove that our strategy is sound, complete, and optimal, modulo non-deterministic choices, for appropriate definitions of these concepts. We compare our strategy with related work and show that our overall approach is fully compatible with the current proposal of a universal, broad-based functional logic language.