Retrievals of microwave land emissivity have been performed for three summer seasons (June--August) over the conterminous United States for the years of 2000, 2001, and 2002. These retrievals use observations from the Special Sensor Microwave Imager to construct a database from which means and channel correlations of land emissivity can be computed, needed for variational retrievals of atmospheric parameters. In addition, the study establishes a range of values that must be reproducible by forward emissivity models. Embedded in the retrieval is an infrared-based cloud-clearing algorithm and land surface temperature retrieval. The land surface temperature retrieval employs vegetation and soil databases to index infrared spectral reflectance measurements to a given location. Analysis reveals that the land surface temperature dominates the total error in microwave emissivity, with the largest contribution coming from impacts of subpixel and undetected clouds, and interpolation error of 3-hourly land surface temperature to the time of the satellite overpass. A practical microwave emissivity error estimate is 2% for all frequencies less than 85 GHz, within a natural summertime variability of 5%. A clearly polarized signature from the surface is present in the 19 GHz frequencies over terrain characterized as predominantly bare ground, with the scattering by vegetation and small-scale roughness effects reducing this polarized signal at higher frequencies and over more densely vegetated areas.