Abstract: Using radar measurements from different frequency bands which are distributed sparsely in one dimensional spectrum, multi-band radar signal fusion can improve accuracy of estimation of radar target scattering model parameters and range resolution of the range profile by signal level’s coherent fusion. The performance of traditional fusion based on spectrum estimation is limited by estimation of scattering model order. Furthermore, owing to sparse distribution of multi-band, the mutual coherence of observation system matrix is destroyed, and the global optimal solution of Basis Pursuit (sparse representation based on l1-norm) may be unequal to real sparse representation of signal. Thus a new method of multi-band radar signal fusion based on sparse Bayesian learning is proposed in this paper based on GTD model. This method avoids the step of the model order estimation, and overcomes the limitation of Basis Pursuit in multi-band signal fusion. The experimental results also show the advantage of this method.