This paper shows how to extend an existing Prolog system to automatically exploit OR-parallelism. The description starts with parallelizing pure Prolog, a Prolog version without either cut or side effects. The model is incrementally refined until finally reaching an efficient OR-parallel system for full Prolog with extra non-Prolog features. I have tried to keep the text as general as possible. When the text becomes too SICStus (my target Prolog system) specific some hints for the ordinary WAM is given. The chosen OR-parallel model is the Muse model. It is relatively straightforward using this model to extend most existing Prolog systems.