Abstract: In data mining reseach, the classification algorithms generally pursue more highly accuracy. It is based on the assumption that all misclassifications have the same cost. However, the assumption is not correct in the real world, so that the normal classification algorithms do not perform well. By improving the encode/decode methods and taking different misclassification cost into account, this paper concerns a new cost-sensitive algorithm called CSC-GEP based on Gene Expression Programming (GEP). The experimental results show that the new algorithm is effective.