Quantum state discrimination has been an important topic in quantum information theory. In particular, a locally separated party performing such a task sometimes leads to counter-intuitive results, including nonlocality without entanglement and LOCC discrimination of any two orthogonal states. In this talk, we study a LOCC discrimination task of a set of maximally entangled states by imposing an additional condition that the state should not be disturbed after the measurement. We demonstrate that the best strategy becomes a random guessing without pre-shared entanglement, while the probability for successful discrimination reaches 1 with preshared Bell states. We further discuss the entanglement cost associated with non-destructive quantum state discrimination.