Diurnal temperature range (DTR), usually defined by difference between daily maximum temperature and daily minimum temperature, is an important climate metric that can provide insight about mechanisms regulating contemporary change in energy budgets of the atmosphere and land surface. Decreases in DTR emerge in the observational record over the past half century, with ESM often underestimating the observed magnitude of the decline. Here we used CRU observations and CMIP5 and CMIP6 model simulations to explore drivers of DTR trends. A strong downward trend in DTR during 1960-2016 is visible in most regions in observations. However many ESM simulations underestimate the magnitude of the observed decline. While increases in minimum temperature are faster than increases in maximum temperature in observations, in many models this pattern is reversed, particularly during boreal summer. We are able to explain some of spatial structure of observed and modeled DTR trends using trends in cloud fraction and relative humidity. In many areas, neutral to positive cloud and relative humidity trends occur in observations, whereas negative trends often occur in the models. We hypothesize that a strong physiological response to rising CO₂ in the models reduces near surface humidity, allowing for greater nighttime longwave radiation losses and causing the models to under predict the negative DTR trend. To explore this hypothesis, we analyzed idealized climate simulations that isolated the radiative effects of CO₂ (esmFdbk1) from CO₂-physiology effects (esmFixClim1). We find that the radiative effects of CO₂ have a dominant impact on the DTR trend in the Northern Hemisphere during winter, whereas the physiological effects of CO₂ are dominant during summer. These results suggest that the response of evapotranspiration and near surface humidity to rising CO₂ may be too strong in the models, and as a consequence, they underestimate the declining DTR trend. Apart from the low bias in DTR trend within the models, the climatology of DTR is also underestimated. We integrated several DTR metrics into ILAMBv2 software to aid with future development aimed at reducing these biases. Further investigation reveals that low bias in DTR climatology may relate to positive biases in high cloud fraction and relative humidity as simulated by the ESMs.