An atmospheric general circulation model is coupled with three different land surface schemes (LSSs) to study the impact of LSS uncertainties on the simulation of surface warming caused by doubled CO2 and associated oceanic changes. It is found that the implemented LSSs can greatly impact the spatial distribution and amplitude of the simulated warming over land. In warm regions, the different descriptions of local land processes and their associated feedbacks are responsible for about half of the inter‐model spread. In cold regions, almost all of the spread is caused by the circulation differences triggered by LSS differences, and the local land processes have little direct impact. On the other hand, as the annual global‐average land surface warmings are very close no matter which LSS is coupled, this study suggests that the LSS has little impact on projections of annual global‐average temperature change, despite strong regional impact in warm regions. Copyright © 2010 Royal Meteorological Society