Based on an analysis of the difference between EEMD and wavelet transform, this study investigates the multi-scale characteristics of the precipitation and temperature in Guangzhou. The results are presented as follows. (1)Wavelet analysis is non-adaptive and may cause spurious oscillations, since the convolution of a priori basis function and a priori number of levels are variable. (2) EEMD method, which decomposes non-linear and non-stationary signals into basis functions (called the intrinsic mode functions) adaptively, ensures the stability of results. The ensemble results with different amplitude white noise are more consistent, which demonstrates the robustness and advantage of EEMD over wavelet analysis. (3) The results from EEMD analysis show that there are mainly 5 time-scale fluctuations in temperature and precipitation in Guangzhou from 1908 to 2010. For temperature and precipitation, the interannual scales are 2~6.8 a. On the interdecadal scales, temperature is of 13.6 a, 34 a, 68 a and precipitation 22.6 a.