Abstract: OBJECTIVE To distinguish Anoectochilus roxburghii and relative species by near infrared(NIR) spectroscopy combined with chemometrics, and to establish a prediction model for rapid determine polysaccharides contents in Anoectochilus roxburghii. METHODS The NIR spectroscopy of Anoectochilus roxburghii, Anoectochilus formosanus Hayata and Ludisia discolor were collected. The prepossessing of original spectrum was optimization through accuracy of classification in the NIR model, and six supervised pattern recognition algorithms such as decision tree, K-nearest neighbor algorithm, random forest, partial least squares regression discriminant analysis, linear discriminant analysis and support vector machine(SVM) were applied to identify effect of the classification effect, optimum algorithm and then establish qualitative model. The content of polysaccharides in 76 batches of Anoectochilus roxburghii samples were examined by ultraviolet visible spectrophotometry combined with phenol sulfuric acid method. In order to select optimization algorithm, six quantitative stoichiometry algorithms consisted of SVM, extreme learning machines, decision trees, random forests, principal component regression and partial least squares regression(PLS) were used to connect polysaccharide content and the NIR spectroscopy in Anoectochilus roxburghii respectively. The best method for determining the content of Anoectochilus roxburghii polysaccharides was further optimized by spectra pretreatment, band selection and number of band variables based on successive projection algorithm(SPA). RESULTS The NIR discriminant analysis model was established by SVM with SNV+SG+2^ndD, and the classification accuracy was best. The prediction performance was evaluated based on the radial basis kernel function algorithm combined with confusion matrix and ROC curve, and the model performance was good. In addition, the quantitative analysis model was constructed by continuous projection-partial least squares by the prepossessing of SNV+SG+2^ndD and the optimal band of 7 000-4 000 cm^-1 with 97 of variables number. The accuracy was 0.992, which was the highest. The root mean square error calibration set, correlation coefficient of calibration set, and the root mean square error in validation set, correlation coefficient of validation set were 0.625, 0.993, 0.767, 0.992, separately. The prediction deviation was 8.467, and relative deviation of prediction set was less than 10%. CONCLUSION The established NIR-SVM qualitative model and SPA-PLS quantitative model are accurate and reliable, which are enable to identify Anoectochilus roxburghii and determine polysaccharide content nondestructively. It is a new and promising method for rapid evaluation of Anoectochilus roxburghii quality.