A new efficient method to obtain high resolution ISAR images in the case of missing observation data is presented. The available data segments are modeled by 2-D linear prediction of 2-D Cartesian frequency spectra using 2-D orthogonal lattice filters. Then the prediction models are used to estimate the missing data segments. It is shown that the EFT processing of the resulting data achieves better resolved images both in range and cross-range. Our algorithm is based on the 2-D prediction of the backscattered data, while other existing algorithms model range or cross-range profiles separately. Therefore, our method provides a better modeling for the gapped data.