The impact of transaction costs on the early emissions trading market is examined by applying an agent based model and simulation (ABMS) approach. For a realistic model set up, bounded rationality, stochastic characteristics, and learning-by-doing are considered in our search processes. Marginal abatement cost parameters are obtained from Yoo et al. (2010), which is an experimental study on the emissions trading in the Korean power sector. Sensitivity analyses are performed on market performance indices with regard to transaction cost parameters, which represent scales and the learning elasticities of transaction costs. A total of 960 simulations were run in this sensitivity analysis. Sensitivity analysis results consistently show that higher transaction costs worsen market performance. The most remarkable finding in these results is that welfare performance of all the transactions decreases by up to 50% as the scale parameters of transaction costs increase, implying that welfare gain from introducing emissions trading disappears significantly. However, with learning curve effect, welfare performance could be regained by up to 26%. In sum, although transaction costs significantly encroach upon trade gains at the early stage, based on our simulation results, the welfare loss by way of transaction costs is lessened as the knowledge of market participants progresses. (C) 2015 Elsevier Ltd. All rights reserved.