In this work, we investigated the idea of identifying DNA-binding proteins from sequence by combining SVM and PSSM-DT. The PSSM-DT is the features from PSSM by considering the probabilities of pairs of amino acid separated by certain number of sites along the sequence in a sequence. A benchmark test on a dataset of 525 DNA-binding proteins and 550 proteins which do not bind to DNA using jackknife validation showed that SVM-PSSM-DT achieved the best predicting performance with ACC of 79.96%, MCC of 0.62 and AUC of 86.50%, and performed better than other state-of-the-art methods by 4.56-7.41% in terms of ACC, 5-10.4% in terms of AUC and 0.12-0.18 in terms of MCC. Subsequently, the blind test performed on the Independent dataset PDB186 indicated that the proposed predictive method obtain an ACC of 80.00%, MCC of 0.647 and AUC of 87.40%, and outperformed some existing state-of-the-art methods. Additionally, the discriminant weight of the descriptors in PSSM-DT-based protein representation is calculated based on the benchmark dataset and the analysis results show that pair(R, R), pair(R, P), pair(P, R) and pair(A, R) are the top four most discriminative amino acid pairs. The three dimensional structure of the protein 1AKH chain A showed that the descriptors for the top four most discriminative amino acid pairs only occur in the DNA-binding regions of the protein, indicating that PSSM-DT is a useful tool for identifying DNA-binding protein.