The convergence of economic growth is the key prediction of the neoclassical growth theory, but the related estimation of domestic references on convergence speed is generally high. In this paper, by carefully selecting the control variables and considering the spatial correlation and the co-linearity which may exist among the control variables, we estimate the spatial panel model based on the extraction of principal components, the obtained convergence speed is only 2.36%. The result is not only closer to the experience value of foreign literature, but also more suitable for our economic reality. This tells us, on the convergence mechanism of economic growth, we should not only select the control variables carefully, but also select the estimation method of the model properly.