Integrated assessment models (IAMs) have commonly been used to understand the relationship between the economy, the earth's climate system and climate impacts. We compare the IPCC simulations of CO2 concentration, radiative forcing, and global mean temperature changes associated with five SRES 'marker' emissions scenarios with the responses of three IAMs-DICE, FUND and PAGE-to these same emission scenarios. We also compare differences in simulated temperature increase resulting from moving from a high to a low emissions scenario. These IAMs offer a range of climate outcomes, some of which are inconsistent with those of IPCC, due to differing treatments of the carbon cycle and of the temperature response to radiative forcing. In particular, in FUND temperatures up until 2100 are relatively similar for the four emissions scenarios, and temperature reductions upon switching to lower emissions scenarios are small. PAGE incorporates strong carbon cycle feedbacks, leading to higher CO2 concentrations in the twenty-second century than other models. Such IAMs are frequently applied to determine 'optimal' climate policy in a cost-benefit approach. Models such as FUND which show smaller temperature responses to reducing emissions than IPCC simulations on comparable timescales will underestimate the benefits of emission reductions and hence the calculated 'optimal' level of investment in mitigation.