A novel approach for identification of flow pattern of micro-channel nanofluid gas—liquid two-phase flow was presented based on K-means for the purpose of improving the accuracy and efficiency of flow patterns identification. The proposed flow pattern identification method acquired the whole flow pattern images of the gas—liquid two-phase flow of micro-channel with high-speed camera firstly. In the second place, peak values which were obtained by histogram of gray scale, flow pattern images were thought of as the original center point of K-means clustering. As for the final step, similarity identification of different flow pattern images was carried out with the principles of invariant moment theory and Euclidean distance. The accuracy and efficiency of the proposed flow pattern identification method were demonstrated with the precision-ratio and recall-ratio assessment system as well as time-consuming analysis results of fifty five hundred pieces of flow pattern images identification experiment. Experimental results showed that the overall identification rate of the new flow pattern identification method based on K-means clustering was 97.8%, while the identification rate of slug flow was up to 100% and that of bubble flow was able to reach 100% as well. The new method provided a novel perspective for the online identification of flow pattern of micro-channel nanofluid two-phase flow.