The Swiss National Centre of Competence in Research on Climate aims to explore the predictability, variability, and risk of climate change. The Paul Scherrer Institute, which is involved in this program, uses integrated assessment models to simulate policies for climate-change mitigation under uncertainty. We report here selected results of the model for evaluating the regional and global effects (MERGEs) of greenhouse-gas emissions with endogenous technological learning (ETL), known as MERGE-ETL. The novelty of the approach is the application of an heuristic algorithm to solve the non-linear and non-convex MERGE problem where 'learning-by-doing' is adopted for a set of energy technologies. The study presents numerical examples showing the implications of endogenous-learning for the timing of carbon-abatement that stabilizes carbon concentrations (e.g. at 550 ppmv), as well as the implications of this in terms of cost/benefit (C/B) analysis. The endogenous-learning formulation is contrasted with the version of the model without ETL. The improved methodology indicates a potential for significant reduction in carbon-abatement cost and economic losses. The method, which is basically in favor of late actions in abatement, implicitly assumes early R&D support and learning investments in carbon-free systems to help these new technologies follow their learning curves. The endogenous treatment of learning already shows significant reductions of carbon emissions in the baseline case and indicates that low-carbon concentrations and improved environmental performance can be obtained when policies are followed that compensate for externalities related to climate change. More precisely, MERGE-ETL gives C/B-optimal carbon-emission trajectories near the 590-ppmv-concentration level. Moreover, the imposition of constraints on the rate of temperature change (e.g. 0.21 degrees C per decade) demands early actions in carbon mitigation. (c) 2004 Elsevier Ltd. All rights reserved.