In this paper, a four-stage ensemble support vectormachine (ESVM) based on multi-agent learning approach is proposed for credit rating system in electronic commerce. In the first stage, the initial credit dataset is divided into two independent subsets: training credit subset (in-sample data) and testing credit subset (out-of-sample data) for training and verification purposes. In the second stage, different ESVM learning paradigms with much dissimilarity are constructed as intelligent agents for credit rating evaluation. In the third stage, multiple individual ESVMagents are trained using training rating subsets and the corresponding rating results are also obtained. In the final stage, all individual rating results produced by ESVMin the previous stage are aggregated into an ensemble rating result. In particular, the impact of the diversity of individual intelligent agents on the generalization performance of the ESVM-based multi-agent learning way is examined and analyzed. For illustration, one corporate credit rating dataset is used to verify the effectiveness of the ESVM-based multiagent learning system.