Euclidean distance optimized antenna selection (EDAS) can significantly improve the bit-error-rate (BER) performance of spatial modulation (SM) systems. However, the exhaustive search over all possible antenna subsets leads to inherently high-search complexity. In this letter, a cross-entropy-based antenna selection (CEAS) scheme is proposed, by formulating the antenna selection problem in SM as a combinatorial optimization one. Simulation results show that the proposed CEAS scheme achieves considerable reduction in search complexity, while approaching the optimal performance of exhaustive search.