Abstract: As a result of the acquisition of environment and instrument performance constraints,seismic data collected are often irregular and incomplete. Thus it is necessary to reconstruct the complete seismic data before proceeding to the next step in the seismic data analysis. The authors present a modified Curvelet algorithm for image reconstruction based on seismic compression. First in the framework of compressed sensing theory,using the sparse characteristic of Curvelet,a missing data reconstruction model is built,and then using the CRSI(Curvelet Recovery by sparsity-promo- ting Inversion,CRSI) algorithm framework,adopting improved exponential threshold algorithm,the missing seismic da- ta are restored and reconstructed. In this paper,the authors use four level homogeneous medium model and the seismic data simulated by Marmousi model to carry out numerical experiments of random sparse sampling and reconstruction. The result of the experiment shows that compared with the traditional recon-struction algorithm,the proposed method not only accelerates the convergence speed of the original algorithm,but also guarantees a high SNR of the reconstruc- ted data,which verifies the feasibility and effectiveness of the proposed method.