One year of global surface wind products from multiple operational numerical weather prediction (NWP) forecast/analysis systems are used as comparison data to derive an empirical wind speed model function for the Geosat altimeter. The resulting model function is nearly identical to the modified Chelton-Wentz model for wind speeds from 4.5 to 15 m/s. The large quantity of comparison data at each wind speed allows detailed examination of the data scatter (and hence, model function uncertainty) as a function of wind speed. Highly skewed distributions at low wind speeds are consistent with specular reflections and antenna mispointing errors hypothesized by others on the basis of extremely limited data. Mesoscale variability in the wind field and synoptic scale errors in the NWP products are shown to account for ∼ 30% of the observed scatter of σ0 at each wind speed. The remaining scatter is largest at low winds, and decreases to a nearly constant value of ∼ 12% at speeds greater than 7 m/s. Model function uncertainty expressed more traditionally in units of wind speed is examined for historical model functions as well as the present NWP-based model. The historical models have significant biases at high wind speeds owing to the lack of comparison in situ data used in their construction. Although the narrow swath, sensitivity to mispointing errors at low wind speeds, and small variation of backscatter with winds at high speeds limit the scientific utility of radar altimeter wind measurements, the present study demonstrates that modern operational NWP surface wind products are sufficiently accurate to allow development of fully empirical model functions and associated error analyses.