Objective In the study and practice of the tongue characterization of TCM, experienced doctors found that a large number of the tongue images collected by tongue image instrument don’t meet the clinical requirement, which will directly affects the final result of tongue image analysis. This paper focuses on the research about tongue image quality assessment to provide help for the accurate selection of image in clinic. Methods We analyze the requirements of tongue inspection, statistics of local normalized luminance based on natural scene statistics model, texture, color and geometric features of tongue images are extracted respectively, and then support vector machine (SVM) is used for classification. By the classification accuracy to verify whether this method can provide help for the accurate selection of image in clinic. Results We can get a better evaluation of tongue image quality and pick out the tongue images that used in clinic accurately through the extracted feature of tongue image. Conclusions Experimental results show that the feasibility of quality assessment for tongue image. This approach is expected to be applied to the next generation of tongue image analysis instrument, providing high-quality reference data for assisted tongue image analysis.