The micro motion target image of SAR can be used in radar target recognition since it can reflect the attribute of target information. Based on the fact that the SAR echo of the micro motion target is sparse, the sparse signal representation was established under an over-complete dictionary. A new sparse Bayesian learning named expansion-compression variance-component based method was employed, which only assigns the distinct variance components to the significant signal elements. In addition, the expansion-compression variance-component based method has much less parameters. The imaging results of SAR micro motion target can estimate the micro motion parameter better, and achieve the fine SAR micro motion image under the low signal to noise ratio.