Tree-ring reconstructions of temperature often target trees at altitudinal or latitudinal tree line where annual growth is broadly expected to be limited by and respond to temperature variability. Based on this principal, regions with sparse tree line would seem to be restricted in their potential to reconstruct past temperatures. In the northeastern United States, there are only two published temperature reconstructions. Previous work in the region reconstructing moisture availability, however, has shown that using a greater diversity of species can improve reconstruction model skill. Here, we use a network of 228 tree-ring records composed of 29 species to test the hypothesis that an increase in species diversity among the pool of predictors improves reconstructions of past temperatures. Chamaecyparis thyoides alone explained 31% of the variability in observed cool-season minimum temperatures, but a multispecies model increased the explained variance to 44%. Liriodendron tulipifera, a species not previously used for temperature reconstructions, explained a similar amount of variance as Chamaecyparis thyoides (12.9% and 20.8%, respectively). Increasing the species diversity of tree proxies has the potential for improving reconstruction of paleotemperatures in regions lacking latitudinal or elevational tree lines provided that long-lived hardwood records can be located.