Focusing on the issue that some portions of seismic waveforms were lost due to the mechanical failure of recording devices and damage of seismograms, an inpainting method based on Consistent Sparse Representation (CSR) model was proposed. Firstly, each seismic frame was represented individually in Sparse Representation (SR) model. Secondly, the consistency between spectra of seismic frames was employed. Principal Component Analysis (PCA) method was introduced to extract this consistency between seismic frames. Finally, combining the sparseness of each seismic frame and the consistency between seismic frames, the proposed algorithm was used to inpaint the lost portion. The simulation experiments showed that when the missing duration interval was 50% of the frame length, the inpainting method based on traditional sparse representation model obtained incorrect results whereas the inpainting method based on consistent sparse representation model recovered the lost portion well. The simulation experiments indicate that the performance of the consistent sparse representation inpainting method is much better than the traditional sparse representation inpainting method.