Abstract Aim In this paper, we explore the relationship between pollen concentration in surface soil and extant perennial vegetation across a rapidly urbanizing arid ecosystem along with its surrounding agricultural and native Sonoran Desert land. We also investigate whether taxa behave as expected from known pollination characteristics (e.g. anemophilous vs. zoophilous) and whether the pollen–plant relationship differs between the undeveloped desert and agro-urban regions. Location We used a probability-based sampling scheme at 200, 900-m2 sites across the Central Arizona–Phoenix region, a 6400-km2 area of the south-western USA. Methods Pollen–plant abundances were mapped across the study area using interpolation techniques, summary bar charts and scatterplots, then two statistical approaches were applied to examine the data quantitatively. Firstly, we used regression analysis of paired pollen concentration and plant cover data; secondly,pc-ordwas used on a cross-matrix (contingency table) containing the presence/absence scores of both pollen and plants for each taxon at each site. Results Mapping and quantitative analysis revealed that pollen–plant relationships do vary both between anemophilous and zoophilous groupings and by individual taxon. They also revealed that distribution patterns of both pollen and plant abundance vary consistently across the three main landscape types (desert, urban, agricultural) and that pollen taxa groupings are consistent with pollination method and source plant distribution regardless of landscape type. Main conclusions These findings provide a broad range of comparative data to facilitate palynological reconstruction of past vegetation and aid in assessing types of prehistoric impacts on this vegetation. They also document that significant variation can occur in pollen rain across a city, even for taxa with widely distributed pollen, leading to the conclusion that forecasting pollen distribution patterns for allergy-related pollen types using only one or a few pollen traps is likely to entail substantial error.