Abstract: This study investigates Sonoran Desert plant communities in the Central Arizona - Phoenix Long Term Ecological Research (CAP-LTER) site located in and around metropolitan Phoenix, Arizona. There are two main emphases: (1) an examination of vegetation within undeveloped remnant habitat islands with regard to species richness, nestedness, and species accumulation with area, and (2) an effort to generate maps depicting the distribution of natural vegetation types on desert lands using remotely sensed data. Island-level woody species richness is positively related to island area; this relationship arises from larger islands containing both more individuals and higher elevation environments. Local-scale woody species richness is not influenced by island area, but is structured by passive sampling dependent on plant density, productivity associated with elevation, study site identity, and proportional sampling from the island species pool. Nestedness in woody vegetation arises as a consequence of an aggregate response of constituent species involving multiple mechanisms. Nestedness in herbaceous communities arises from an area effect, involving either extinction or passive sampling, and is reinforced by colonization of exotic taxa. In terms of species-area curves, sample curves in both woody and herbaceous vegetation are most often best fit by sigmoid functions, whereas convex functions best describe the relationship between island area and island species richness. Landsat ETM data was used to generate vegetation maps for subsets of the CAP-LTER, with classes determined from field data collected within the study area. Results were varied, with vegetation on clayey soils mapped to an accuracy of 91%. Other subset maps were 70% accurate or less. Mapping of desert vegetation is particularly challenging since bare soil exposure is high and background soil spectra potentially interfere with vegetation signatures. These results demonstrate that image classification of desert vegetation using only Landsat ETM data can be problematic and may not be practical without other supporting data, such as radar imaging, which generally agrees with results of other efforts.