Symmetry is an important feature of many constraint programs. We show that any problem symmetry acting on a set of symmetry breaking constraints can be used to break symmetry. Different symmetries pick out different solutions in each symmetry class. This simple but powerful idea can be used in a number of different ways. We describe one application within model restarts, a search technique designed to reduce the conflict between symmetry breaking and the branching heuristic. In model restarts, we restart search periodically with a random symmetry of the symmetry breaking constraints. Experimental results show that this symmetry breaking technique is effective in practice on some standard benchmark problems.