This paper employs a two-stage residential sorting model to examine climate change impacts on residential location choices in the US. The estimated coefficients are used to simulate population changes and US migration patterns across regions under hypothetical changes in climate. The main dataset used for estimation is the Integrated Public Use Microdata Sample (IPUMS), which provides demographic characteristics of approximately 2.4 million households located in 283 Metropolitan Statistical Areas (MSAs) of the US in the year 2000. Projected climate data (i.e. extreme temperatures) used for simulation are obtained from the North American Regional Climate Change Assessment Program (NARCCAP). In the estimation component, a two-stage random utility sorting model (RUM) is employed. The first-stage discrete choice model employs a multinomial logit specification to recover heterogeneous parameters associated with MSA specific variables, migration costs, along with the mean indirect utility of each MSA. In particular, the interaction terms of temperature extremes and individual-specific characteristics, such as one’s birth region, age and educational attainment, are used to recover valuations of temperature extremes for different classes of people with potentially different preferences. The second stage of this model decomposes the mean indirect utility obtained from the first stage into its MSA-specific attributes controlling for unobservables using region fixed effects. Migration costs are statistically significant. If migration costs are high, individuals are less likely to relocate for the sake of moderate changes in weather extremes. In the simulation component, the estimated coefficients are used to simulate population changes across regions in the US under hypothetical changes in extreme temperatures. We find that extreme temperature and extreme precipitation reduce utility, and people’s preferences for temperature extremes are heterogeneous. The climate of one’s place of birth and demographic characteristics such as age and educational attainment, are significant factors that lead to preference heterogeneity. In addition, we find that population share in the Southern region and California drop, while population share in Northeastern region increases under hypothetical changes in climate.