Using the daily observational data of temperature, pressure, relative humidity and precipitation in 752 stations of China from 1961 to 2008, three methods of division of season, i.e. Zhang’s method, non-linear similarity method and Fisher optimal dissection method are applied to different meteorological elements fields of China. Results indicate that division of season through multi-variables methods is more reasonable than single variable ones, especially in non-linear similarity method and Fisher optimal dissection method. When considering the seasonal climate transition between successive seasons, the results are in good accordance with actual observations. As a result we recommend the non-linear similarity method as a better way of season division.