Climate change research relies largely on macroscale climate estimates to infer past and predict future effects on organisms. But organisms function at smaller scales and the microclimates they experience are likely far more important in determining effects of and responses to climate change. Empirical characterization of microclimates is rare and unrealistic across large spatial scales (at least in the near term). Mechanistic models have emerged as a promising alternative approach to predicting microclimates by incorporating macroclimatic measurements into biophysical models of heat and mass exchange. The broad utility of these models depends on verifying that they work: that is, testing whether predicted microclimates match measured ones. We compared high frequency measurements of soil temperatures (0-50 cm depth) at 287 sites across North America with temperatures predicted by NicheMapR, a biophysical model that incorporates estimates of air temperature and pressure, wind speed, humidity, solar radiation, and rainfall to predict soil temperatures (to 50 cm depth). Model predictions closely matched measured soil temperatures only in low elevation, flat areas with minimal seasonal temperature variation (61% of sites). For areas at higher elevations, with more complex topography, and with higher seasonal fluctuations in temperature, model predictions were often far from measured temperatures (as much as 10 °C cooler in summer months). These and other mechanistic models will be most useful only when they accurately characterize microclimates experienced by organisms. Better predictions of soil temperature in diverse locations may require consideration of additional factors, including slope steepness, cold air pooling, and cloud cover.