Conflict resolution is a research topic for game theory (GT) and conflict analysis. A decision support system (DSS) is very helpful for conflict decision making. Reinforcement learning (RL) is an efficient method to learn knowledge by agents themselves. Although successful applications of RL have been reported in single-agent domain, a lot of work should be done in the case of multi-agent domain. Nash Q-learning is a famous learning algorithm for multi-agent RL. Based on the Nash Q-learning, a novel DSS: multi-agent RL based DSS (MRLDSS) is proposed in this paper and is tested by using several typical examples of conflict resolution. Experimental results show that the proposed architecture and algorithm can solve conflict resolution problems correctly and efficiently.