Error correcting output codes(ECOC) is an effective way to solve multiclass classification problems. Encoding based on data is one of important methods to having coding matrix. Thereforem, an adaptively encoding method based on subclass and particle swarm optimization(PSO) is proposed. Firstly, the similarity between each pair of patterns is measured by using the confusion matrix, and classes are regrouped based on rules. Then binary partitions are gotten based on pattern combination. Finally, the PSO algorithm is introduced to find the most suitable thresholds, thus obtaining a data driven coding matrix. Experimental results show that the proposed method can provide better performance.