Abstract: The long-term control simulation (20th century experiment) of four versions of the coupled atmosphere-ocean global climate model FGOALS (Flexible Global Ocean-Atmosphere-Land System Model), g2.0, s2.0, g1.1 and g1, developed at the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics (LASG), Institute of Atmospheric Physics (IAP), Chinese Academy of Sciences, combined with observational and reanalysis data, are analyzed and compared. In terms of evaluating how realistic the model is in simulating the climatology, the physical mechanisms of Pacific decadal variations are discussed. It is found that g2.0 and s2.0 simulate more reasonable interdecadal Pacific variations than g1.1 and g1. The results indicate that the positive air-sea interaction feedback should be related with the Bjerknes positive feedback mechanism. The negative feedback is mainly associated with the ocean dynamic process. The anomalous meridional heat transport in the Pacific links the tropical and extratropical regions, which inhibits the positive feedback effect. The tropical SST (Sea Surface Temperature) warm anomaly signal, via the atmospheric bridge, affects the extratropical atmospheric circulation. With the air-sea interaction, Kelvin waves and Rossby waves spread across the tropical and extratropical subsurface ocean, respectively. Such negative feedback behavior can be reproduced by the four versions of FGOALS. The study quantifies some of the factors responsible for the periods of Pacific interdecadal variations. The farther from the equator the maximum zonal SST is, the wider the decadal variations are. Subsequently, poleward heat transport requires a longer time, the surface wind stress curl anomalies will extend farther north, and ocean Rossby waves appear much more northward with longer transport periods.