For a better understanding about the protein interactions, the signal transmission direction between protein interactions is used to annotate the protein interaction network. A method based on the physicochemical properties of the domain is proposed to predict the protein interaction direction. Firstly, the physicochemical properties of protein domains is extracted to form the eigenvectors representing the direction information. Then a support vector machine prediction model can be established, and the model is optimized by a grid search method. Finally, the prediction is performed with this model and the optimized parameters. The experimental results show that the accuracy of the model is 88.17% and the AUC value is 0.837. Compared with the PIDS method, this method with the ten physicochemical properties of the protein domain is more efficient for the prediction of protein interaction direction. This work provides an effective new way to predict the protein interaction direction.