The Empirical Orthogonal Function (EOF) analysis was applied to study the thermospheric total mass density at the altitude of 400 km with the data obtained from German CHAMP satellite during the interval from 2003 to 2007, when the solar activity is in declining period. The solar cycle and yearly variations of the thermospheric total mass density ρ are obtained. It is found that the total mass density ρ was obviously affected by the solar activity, the correlation coefficient between solar cycle variation component of ρ and F10.7 index can reach 94.5%. At the mid- and high-latitudes, the amplitude of solar cycle variation is larger in the south hemisphere than in the north. The amplitude of the solar cycle variation is decreased as the latitudes change from low to high. At low latitudes there exists a structure of double crests which is recently known as the Equatorial Mass density Anomaly (EMA). In the yearly variation of the total mass density, there is a seasonal variation that the density is larger in summer than in winter. The amplitude of the yearly variation of density ρ increases with solar F10.7 index as well as the absolute latitude. Comparing the result of CHAMP data with the output of NRLMSISE00 model under the input of the same condition of observation, it is found that both the solar cycle and the yearly variations of thermospheric total mass density are in good agreement. Even though, the latitude difference of the solar cycle variation of ρ is a little bigger from NRLMSISE00 than CHAMP data and there is no EMA structure in NRLMSISE00 model. Furthermore, there is an obvious semiannual variation component while analyzing the NRLMSISE00 data without 130-day-wide sliding window. Thus it is considered that, with the restriction of CHAMP orbit, the 130-day-wide sliding window may smooth out the semiannual component, as well as the higher. It is concluded that the present results are meaningful in the study of the thermospheric climatology.