We automatically verify Back's atomicity refinement law and a classical data refinement law for action systems. Our novel approach mechanises a refinement calculus based on Kleene algebras in an off the shelf resolution and paramodulation theorem prover and a counterexample checker with heuristics for hypothesis learning. The proofs are supported by a toolkit of meaningful refinement laws that has also been verified and that, for the first time, allows the refinement of programs and software systems, and the verification of further complex refinement laws, by automated deduction. This suggests that a substantial proportion of refinement could indeed be automated.