For insufficient sampling data existing in the practical application of Photoacoustic Tomography (PAT), a reconstruction method based on total variation method was proposed for photoacoustic imaging to solve this problem. The residual between the real signals and the simulated ones from the reconstructed image was calculated for the update image, and the iteration was implemented to obtain the reconstruction image. During the process of iteration, the total variation method in the compressed sensing was utilized to obtain the image with the smallest total variation value by the gradient descent method. Through the numerical simulation, the image reconstruction in the case of insufficient sampling data was accomplished. The results demonstrate that the reconstruction method based on the total variation has better performance as compared with the filtered back-projection method, deconvolution reconstruction method and algebraic reconstruction method. For the 30 sampling points, the peak signal-noise ratio of the reconstructed image is 30.98, 22.09, 8.35 dB higher than those reconstructed by tree kinds of other methods metioned above, respectively. The result of in vitro experiment also shows that this method is more effective for noise suppression.