Spaceborne altimeters are an important data source for obtaining global sea surface wind speeds (U10). Although many altimeter U10 algorithms have been proposed and they perform well, there is still room for improvement. In this study, the data from ten altimeters were collocated with buoys to investigate the error of the altimeter U10 retrievals. The U10 residuals were found to be significantly dependent on many oceanic and atmospheric parameters. Because these oceanic and atmospheric parameters are inter-correlated, an asymptotic strategy was used to isolate the impact of different parameters and establish a neural-network-based correction model of altimeter U10. The results indicated that significant wave heights and mean wave periods are effective in correcting U10 retrievals, probably due to the tilting modulation of long-waves on the sea surface. After the wave correction, the root-mean-square error of the retrieved U10 was reduced from 1.42 m/s to 1.24 m/s and the impacts of thermodynamic parameters, such as sea surface (air) temperate, became negligible. The U10 residuals after correction showed that the atmospheric instability can lead to errors on extrapolated buoy U10. The buoy measurements with large air-sea temperature differences need to be excluded in the Cal/Val of remotely sensed U10.