The key to realize precise chemical application is weed identification. This paper introduces a kind of multi-classification mode based on Support Vector Machines (SVM) and one-against-one-algorithm for weed seedlings (Dchinochloa crasgalli, and Echinochloa crusgalli) in corn and soybean fields. A handheld FieldSpec® 3 Spectroradiometer manufactured by ASD Inc., in USA was used to measure the spectroscopic data of the canopies of the seedlings of corn, soybean, and weeds within 350¡«2500nm wavelength in fields. Each canopy was measured successively and rapidly. The five original spectroscopic data collected over the whole wavelength were averaged in order to eliminate random noise; the averaged original data were converted into reflectance data, and the unsmooth parts of reflectance spectral curves with large noise removed. The effective wavelength range for spectral data processing was selected as 350-1300nm and 1400-1800nm. Two-classification models based on SVM were built up using ‘linear’, ‘polynomial’, ‘RBF’ (Radial Basis Function), and ‘mlp (Multilayer Perception)’ kernels, respectively. Comparison of different kernel functions for SVM shows that higher precision can be obtained by using ‘Polynomial’ kernel function with third-order. The accurate identification rate can be up to 80% and 85%, respectively for corn and soybean, but the SV ratio is relatively lower. On the basis of two-classification SVM model, and by using of one-against-one-algorithm voting procedure, a multi-classification model was set up for corn and soybean in fields. The accurate identification rates reach 80% for corn and 83.3% for soybean. The comparison of three-classification accurate rates for SVM, Neural Network (NN), and Decision Tree (DT) methods was also done using the same sample data measured in corn fields, which are 80% for SVM, 78% for NN, and 63% for DT. SVM achieved the highest correct classification rate.