NASA scatterometer (NSCAT) measurements of the western Pacific Supertyphoon Violet are presented for revolutions 478 and 485 that occurred in September 1996. A tropical cyclone planetary boundary layer numerical model, which uses conventional meteorological and geostationary cloud data, is used to estimate the winds at 10-m elevation in the cyclone. These model winds are then compared with the winds inferred from the NSCAT backscatter data by means of a novel approach that allows a wind speed to be recovered from each individual backscatter cell. This spatial adaptive (wind vector) retrieval algorithm employs several unique steps. The backscatter values are first regrouped in terms of closest neighbors in sets of four. The maximum likelihood estimates of speed and direction are then used to obtain speeds and directions for each group. Since the cyclonic flow around the tropical cyclone is known, NSCAT wind direction alias selection is easily accomplished. The selected wind directions are then used to convert each individual backscatter value to a wind speed. The results are compared to the winds obtained from the tropical cyclone boundary layer model. The NSCAT project baseline geophysical model function, NSCAT 1, was found to yield wind speeds that were systematically too low, even after editing for suspected rain areas of the cyclone. A new geophysical model function was developed using conventional NSCAT data and airborne Ku band scatterometer measurements in an Atlantic hurricane. This new model uses the neural network method and yields substantially better agreement with the winds obtained from the boundary layer model according to the statistical tests that were used. ¿ 1999 American Geophysical Union