This paper uses data from Chinese prefecture-level administrative unit to examine the extent of spatial variability of the impact that population, income, and climate have on urban residential carbon emissions. The residuals of OLS estimation of urban residential carbon emissions exhibit a significant spatial association according to the value of the Moran's I statistic. GWR model effectively reduces the spatial autocorrelation of residuals by considering spatial effect. Not only does it enhance the explanatory power of the model, but also gets local estimates of the parameters. Results show that, there is strong evidence of spatial heterogeneity for impacts of three independent variables: (1) local regression coefficients of population and income are both positive in the OLS and GWR models, but spatial variability of the effect of income is greater in the GWR model; (2) the coefficient estimate of the climate variable in the OLS model is negative, however, the direction is both positive and negative in the GWR model with the magnitude of the effect varying within and across the 302 prefecture-level administrative units in China; (3) one should carefully check the reasonableness of policy recommendations made based on global linear regression models that ignore or failed to properly assess the spatial dependence.