More than a dozen of wind speed (U) algorithms have been proposed during the past 2 decades, as a result of a continuing effort to improve altimeter wind measurement. The progress in terms of accuracy, however, is seen to be rather slow. The reported root mean square (RMS) error of prevailing algorithms varies mostly between 1.6 and 2.0 m/s for the dominant wind regime. As far as the TOPEX altimeter is concerned, three measured quantities, namely, the radar cross sections from Ku and C band (sigma(Ku) and sigma(C)), as well as the significant wave height (H-s), have been used in previous algorithm developments, resulting in a variety of single-, dual-, and three-parameter model functions. On the basis of the finding of a banded dependency of the U-sigma(Ku) relationship on sigma(C) a new approach for retrieving altimeter wind speed, termed linear composite method (LCM), is proposed in this study. The LCM model function appears as a set of sigma(C)-dependent linear relations between U and sigma(Ku). A unique advantage of this approach is that it allows the algorithm to be tuned or expanded for a given range of wind speed without affecting the rest. Over 1.7 million coincident TOPEX/NASA scatterometer (NSCAT) and TOPEX/QuikSCAT data covering a period of 2.5 years are used to adjust the model. Validation against extensive buoy measurements indicates that the LCM algorithm is almost unbiased and has an overall RMS error of 1.56 m/s, which is 12% lower compared to the algorithm in operational use [Witter and Chelton, 1991]. In addition, a small (2.5-6%, depending on the reference data set) but significant improvement is found for the LCM when compared to the most recent dual- parameter algorithm [Gourrion et al., 2002].