Abstract: Traditional frequency agility ISAR imaging method suffers from high sidelobe and low resolution. To improve the resolution, by exploiting the sparsity of targets in the received echo, this paper uses the sparse Bayesian algorithm, namely Expansion-Compression Variance-component based method (ExCoV), to reconstruct the ISAR image from the original Compressed Sensing (CS) ISAR data. By taking into account of the prior information of the sparse signal and the additive noise encountered in the measurement process, the sparse recover algorithm under the Bayesian framework can reconstruct the scatter coefficient better than the traditional methods. Different from the Sparse Bayesian Learning (SBL) endowing variance-components to all elements, the ExCoV only endows variance-components to the significant signal elements. This leads to much less parameters and faster implementation of the ExCoV than the SBL. The simulation results indicate that it can conquer the problem brought by traditional methods and achieve high precision agility ISAR imaging under the low SNR.