We consider the refinement of a static analysis method called thread-modular verification. It was an open question whether such a refinement can be done automatically. We present a counterexample-guided abstraction refinement algorithm for thread-modular verification and demonstrate its potential, both theoretically and practically.