Wind stress variability over the Benguela upwelling system is considered using 16 months (01 August 1999 to 29 November 2000) of satellite-derived QuikSCAT wind data. Variability is investigated using a type of artificial neural network, the self-organizing map (SOM), and a wavelet analysis. The SOM and wavelet analysis are applied to an extracted data set to find that the system may be divided into six discrete wind regimes. The wavelet power spectra for these wind regions span a range of frequencies from 4 to 64 days, with each region appearing to contain distinct periodicities. To the north, 10¿--23.5¿S, the majority of the power occurs during austral winter, with a 4--16 day periodicity. Further investigation of National Centers for Environmental Prediction reanalysis outgoing longwave radiation data indicates that the winter intensification of wind stress off the Angolan coast is linked with convective activity over equatorial West Africa. The summer activity appears to be linked with the intensification of the Angolan heat low. Convective activity over the Congo basin appears to impact upon wind stress variability, off the Angolan coast, throughout the year. Farther south, 24¿--35¿S, the majority of the power occurs in the summer. Here a bimodal distribution occurs, with peaks of 4--12 and 25--50 days. The southernmost regions appear to be forced at higher frequencies by both midlatitude cyclones (austral winter) and mesoscale coastal lows (austral summer). At lower frequencies, eastward propagating periodic wind events that originate over eastern South America appear to be important to the forcing of wind stress over the southern Benguela.