The mapping between segments and control sites is an important factor that influences the correction result of model-based optical proximity correction (MB-OPC). The existing mapping schemes are rule-based and it is either the inaccurate one-to-one mapping or the time-consuming all-to-one mapping.A new MB-OPC with model-based mapping between segments and control sites is proposed. The mapping model is a charasteristic area derived from gradient of intensity,and it can be a one-to-one mapping or a multiple-to-one mapping. The mapping can be applied to all the control sites in MB-OPC,and there is no need to recalculate repeatily. The experimental results show that the variance of edge placement error (EPE) of MB-OPC with model-based mapping is significantly improved and the run time is not significantly increased.