A prediction model has been proposed of drug target interaction based on local and global consistency (LLGC) learning. Based on the conclusion that adjacent nodes and nodes in manifold structures or clusters are more likely to have the same label, a comprehensive consideration is given to the global and local characteristics of target and drug data, with the sequence similarities of targets and topological structure information of drug-target network integrated together, and a consistency learning method is proposed to mine the drug target interaction data from standard data set according to standard data set. In order to analyze the performance of LLGC, a comparison has thus been made between the proposed method and SBGI, KBMF2K, NetCBP and WNN-GIP in four datasets of enzyme, ion channel, GPCR and nuclear receptor. The experimental results show that the performance of LLGC is better than other methods except that the AUC value of LLGC in nuclear receptor data is slightly lower than that of NetCBP and WNN-GIP. After determining the performance of the LLGC model, the model is applied for the drug-target interaction prediction, with the five drug-target interaction data with the highest score thus given. The results show that most of the known drug-target interactions in the four data sets appear in the first 20% of the prediction results, with more than 91% appearing in the first 50% of the prediction set. This result show that LLGC can effectively predict potential associations between drugs and targets.