A neural network inversion scheme including first guess information has been developed to retrieve surface temperature Ts, along with atmospheric water vapor, cloud liquid water, and surface emissivities over land from a combined analysis of Special Sensor Microwave/Imager (SSM/I) and International Satellite Cloud Climatology Project (ISCCP) data. In the absence of routine in situ surface skin measurements, retrieved Ts values are evaluated by comparison to the surface air temperature Tair measured by the meteorological station network. The Ts - Tair difference shows all the expected variations with solar flux, soil characteristics, and cloudiness. During daytime the Ts - Tair difference is driven by the solar insulation, with positive differences that increase with increasing solar flux. With decreasing soil and vegetation moisture the evaporation rate decreases, increasing the sensible heat flux, thus requiring larger Ts - Tair differences. Nighttime Ts - Tair differences are governed by the longwave radiation balance, with Ts usually closer or lower than Tair. The presence of clouds dampens all the difference. After suppression of the variability associated to the diurnal solar flux variations, the Ts and Tair data sets show very good agreement in their synoptic variations, even for cloudy cases, with no bias and a global rms difference of ~2.9 K. This value is an upper limit of the retrieval rms because it includes errors in the in situ data as well as errors related to imperfect time and space collocations between the satellite and in situ measurements.