A new passive damage detection approach is proposed using the reciprocity of the GreenÕs function. This is achieved using a known method to reconstruct the forward and backward time-domain GreenÕs functions between any two detection points in an acoustic diffuse field from the ensemble average of their cross-correlation functions. Damage is detected when the similarity between the forward and backward signals decreases due to the wave nonlinearities introduced by the discontinuity. Proof-of-principle results are discussed for an aluminum plate. A simulated diffuse field was created on the structures using a grid of independent random excitations. Aiming at damage detection, the observations were collected by two sets of experiments considering pristine structure and the case with simulated damages. Due to the lack of exact knowledge to the statistical distribution (most likely in a real world wind turbine blade), a bootstrapping parameter estimation was proposed in this study. To classify the cases with and without damages, a feature study was conducted based on multiple bootstrapping results. This damage detection method combining with the empirical statistical framework can be appropriately extended to any structure with significant acoustic or ultrasonic noise sources.