Abstract: In this paper, the nonlinear theory is used to analysis multi-level climate system.It is indicated that fractal dimension characterizes the structure of climatesystem and is an invariable after scale transformation in the system.Fractal dimension can be used to smooth the climate data, filtering smaller scale fluctuations and forming higher climate level. The structures of climate levels before and after smooth are different. The tendency of climate change in lower level is often not clear but it becomes clearer in higher level. There are nonlinear interactions between higher and lower levels,These are some basic problems in study of climate. And they are investigated in this paper by the use of some climate data sets of China.