Globally distributed crossovers of altimeter and scatterometer observations clearly demonstrate that ocean altimeter backscatter correlates with both the near-surface wind speed and the sea state. Satellite data from TOPEX/Poseidon and NSCAT are used to develop an empirical altimeter wind speed model that attenuates the sea-state signature and improves upon the present operational altimeter wind model. The inversion is defined using a multilayer perceptron neural network with altimeter-derived backscatter and significant wave height as inputs. Comparisons between this new model and past single input routines indicates that the rms wind error is reduced by 10%–15% in tandem with the lowering of wind error residuals dependent on the sea state. Both model intercomparison and validation of the new routine are detailed, including the use of large independent data compilations that include the SeaWinds and ERS scatterometers, ECMWF wind fields, and buoy measurements. The model provides consistent improvement against these varied sources with a wind-independent bias below 0.3 m s21. The continuous form of the defined function, along with the global data used in its derivation, suggest an algorithm suitable for operational application to Ku-band altimeters. Further model improvement through wave height inclusion is limited due to an inherent multivaluedness between any single realization of the altimeter measurement pair [so, HS] and observed near-surface winds. This ambiguity indicates that HS is a limited proxy for variable gravity wave properties that impact upon altimeter backscatter.