This study explored the possibility of using a regression analysis to derive equations to predict differences in hourly air temperatures (DT) between below-canopy urban spaces. During July 1996, data loggers recorded hourly averages of wind speed and direction, air temperature, and relative humidity at about 2 m above ground at three sites in Atlanta, Georgia. The sites were about 1 km apart and differed with respect to tree, grass, and non-vegetative cover in the vicinity. One site was in a small forest in a botanical garden, another was in a backyard in an older residential neighborhood with large trees, and the third was on a small lawn between 3- to 4-story buildings near the downtown central business district. Using the residential site as the reference, we related DT as the dependent variable to differences in descriptors of city morphology and to synoptic conditions as reported for the city airport. The garden-site DT averaged -0.54 C (cooler than the residential site) with a range of -2.3 to +0.8 C, and the downtown site DT averaged 0.57 C (warmer than the residential site) with a range of -0.4 to 1.9 C. The morphology descriptors included tree, grass, and impervious cover for a range of distances up to 8 km in the upwind direction from each of the three sites. These descriptors were derived from analysis of a Landsat TM image and use of an Avenue script in ArcView GIS to derive the land cover differences, which varied significantly with airport wind direction. Descriptors of nearby tree and building structure and solar blockage were obtained from hemispherical photographs taken upward from each site. General weather condition descriptors included estimates of solar radiation inputs and long-wave radiation exchange, atmospheric stability (Turner Class), vapor pressure deficit, and antecedent precipitation over a range of time periods. Multiple regression analysis with seven independent variables yielded equations that explained about 67% of the variance in air temperature differences.