In this paper, four multimodel ensemble methods, namely ensemble mean (EMN), weighted ensemble mean (WEMN), biasremoved ensemble mean (BREM) and weighted biasremoved ensemble mean (superensemble, SUP) have been constructed based on TIGGE datasets, from which four single model forecasts of recent three years from CMA, ECMWF, JMA and NCEP are selected to conduct multimodel ensemble combination and comparison. Single and multimodel ensemble forecasts on the tracks and central pressure of tropical cyclone over Western Pacific with forecast lead time 24 h, 48 h, 72 h, 96 h and 120 h are evaluated and compared. The results show that different single models have different forecast skills. The CMA forecast has the worst performance during the three years, and the best single model forecast in 2010 and 2011 is ECMWF while NCEP is the best one in 2012. Generally, multimodels have significant superiority over the single models within 120 h forecast lead times. As the simplest method within all the four multimodels, EMN shows limited superiority. WEMN has better skill than EMN because it can give different weights to the member models, while the BREM exhibits superiority for the systematic bias is removed. Because of considering both the weights of member models and bias elimination, the SUP has a better skill in general.