Decadal prediction on 10-30 year time scale is one of the most important contents of the 5th phase of the Coupled Model Inter-comparison Project (CMIP5). According to the experiment requirement of CMIP5, a set of decadal experiments were performed using the Beijing Climate Center Climate System Model (BCC-CSM1.1) which is one of models jointed in CMIP5. This study evaluated the model’s prediction capability in regional and global surface temperatures on decadal time scale, and aimed to explore their dependences on the initial observed states of ocean in comparison with the historical experiment in the 20th century using BCC-CSM1.1. The results show as following: (1) BCC_CSM1.1 can simulate the warming trend of 10-year mean global surface temperature not only for oceanic initialization condition but also for without oceanic initialization condition. Nevertheless, the global warming trend simulated by BCC-CSM1.1 can be obviously decreased under the condition of oceanic initialization, which is closer to the observation than that in the historical experiment without oceanic initialization. This feature is much more remarkable in the area between 50°N and 50°S where there are abundant observation data. (2) The nudging method is used to initialize the model with the SODA temperature data. After a “training” period of 8-12 months, predicted surface temperatures in the first year not only in ocean but also in land between 50°S and 50°N are close to CRU observations. Due to the warmer SST bias of SODA reanalysis contrast to HadSST2, there is about a period of 2 to 7 years in decadal experiments that adjusts from the observed ocean state to model basic state. The adjustment time for the ocean and land is almost identical in the same decadal experiment. (3) The prediction skill for decadal-mean SST has strong feature. The high correlations with the CRU observations are mainly near the middle- and high-latitude Indian Ocean in the Southern Hemisphere, the western Pacific Ocean, and the Atlantic Ocean. The oceanic initialization does not significantly influence the prediction results. (4) The variation of decadal-mean predicted SST is closely correlated with the surface heat flux. In the tropical and subtropical region, the net long wave radiation and sensible heating flux has larger influence on the decadal mean SST variation than the net short wave radiation and the latent heating flux, but in oceans at higher latitudes, the variation of decadal mean SST is mostly determined by the latent heating flux.