In this study ,we evaluated the ability of four versions of Flexible Global Ocean-Atmosphere-Land System model (FGOALS) developed at State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics(LASG), Institute of Atmospheric Physics (IAP), Chinese Academy of Sciences in simulating climatology and seasonal cycle of tropical pacific precipitation and SST.FGOALS-f3-L and FGAOLS-g3 participated the Coupled Model Intercomparison Project (CMIP) Phase 6 while FGOALS-s2,FGOALS-g2 participated the CMIP Phase 5 (CMIP5)).This paper compared historical and AMIP simulations in terms of atmosphere-ocean feedback mechanisms and heat budget analysis to investigate the formation of precipitation and SST biases. The results exhibited notable improvements in FGOALS-g3 and FGOALS-f3-L compared with the previous versions ,such as a reduction of 50% of RMSE of SST decreasing from above 2 °C in FGOALS-g2 and FGOALS-s2 to 1 °C in FGOALS-f3-L. However, there are negligible improvements in precipitation. FGOALS-f3-L reproduces the meridional asymmetry of SST and precipitation in East Pacific well because of the reduction of errors in the mean state of shortwave radiation and its reasonable representations of ocean dynamic heat transport and surface shortwave radiation feedback. Moreover, excessive rainfall biases in the northern side of the equator become more severe than previous models, which is greater than 4mm/day. This overestimated convections bring latent heat which enhances the latitudinal ?diabatic heating gradient and strengthen the cross equatorial south wind. Errors in wind actually cancel out part of the warm biases of SST resulting from overestimation of shortwave radiation. Similar error sources existed in the simulated seasonal cycle , characterized by an improved annual cycle but a weaker overall amplitude in FGOALS-g3 and FGOALS-f3-L.This improved annual signal of SST results from stronger amplitude of annual cycle of meridional wind and latent heat flux. Actually, errors of wind and latent heat are bigger but cancel out the annual biases in short wave radiation flux. Therefore, the annual signal in SST seems to be enhanced than former versions. While the false semi-annual cycle of SST in FGOALS-g2 and FGOALS-s2 is mostly result from the semi-annual cycle of latent heat flux.