In this study, the amplitude data from the Annual Bulletin of Chinese Earthquake and the provincial earthquake bulletins of Sichuan and Yunnan were tentatively divided into 6 time segments. Tomographic ML amplitude method has been applied for each time segment to obtain temporal Q value variations. Firstly, a representative time segment with more than 14 000 ray paths was selected to estimate the resolution and error of tomographic attenuation structure by adding random noise to original data and bootstrap method. Then, the same tomographic method with the same parameters was applied for all time segments. The result shows that: 1) the imaging error is small from the observation error of original data and ignoring radiation pattern of earthquake source. The error is less than 6% if 40% random noise is added to the original data; 2) the bootstrap method shows that the maximal error is less than 8% of average Q value, that means the tomographic results are stable and reliable because of huge data set in Sichuan and Yunnan region; 3) in checkerboard test, the resolution can reach 20’ for the region with more than 50 ray paths in each cell; and 4) the Q value variations were obtained after imaging for each time segment. A clear positive relation was found between the average Q value for each time segment and the number of middle and large earthquakes, in other words, the more earthquakes, the larger average Q, that may be resulted from the changes of regional stress.