Rank reduction estimator (RARE) can provide direction-of-arrival (DOA) estimation without knowing the (expectant) array model errors, and avoid iterative computation and local convergence. However, RARE requires the array errors model, and its angle resolution performance is affected by unexpectant model errors in practice. We analyze the angle resolution performance of RARE. The second-order error representation and statistical properties of the perturbed RARE spatial spectrum are obtained based on the derivation of three types of angle resolution probability. The theoretical analysis is validated in the numerical experiment aimed at RARE for mutual coupling self-calibration of uniform array, and direction-dependent amplitude-phase errors self-calibration.