The performance of Automatic Target Recognition (ATR) applications can be increased by enhancing ISAR images which constitute an important step in radar target classification. For this aim, bilateral filter based image decomposition is applied to ISAR images prior to classification. The range and spatial parameteres of Gaussians which constitute the bilateral filter are chosen so as to optimize the classification accuracy rate. While the optimized parameters transfer the speckle to detail bands, they locate the edge information in the residual image. The features obtained from the residual images are applied to several classifiers and the increase in the classification rate is validated for an experimental dataset.