Environmental DNA (eDNA) approaches to monitoring biodiversity in terrestrial environments have largely focused on sampling water bodies, potentially limiting the geographic and taxonomic scope of eDNA investigations. We assessed the performance of two strictly terrestrial eDNA sampling approaches to detect arboreal mammals, a guild with many threatened and poorly studied taxa worldwide, within two central New Jersey (USA) woodlands. We evaluated species detected with metabarcoding using two eDNA collection methods (tree bark vs. soil sampling), and compared the performance of two detection methods (qPCR vs. metabarcoding) within a single species. Our survey, which included 94 sampling events at 21 trees, detected 16 species of mammals, representing over 60% of the diversity expected in the area. More DNA was found for the 8 arboreal versus 8 non-arboreal species detected (mean: 2466 vs. 289 reads/sample). Soil samples revealed a generally similar composition, but a lower diversity, of mammal species. Detection rates for big brown bat were 3.4 × higher for qPCR over metabarcoding, illustrating the enhanced sensitivity of single-species approaches. Our results suggest that sampling eDNA from on and around trees could serve as a useful new monitoring tool for cryptic arboreal mammal communities globally.