Nondestructive testing technology of water-injected meat has developed rapidly, thus a novel method was displayed based on the technical of pattern recognition algorithms, aiming to research a pattern recognition model which combined with spectral analysis technique and support vector machines. Then the examination was designed with the purpose of getting objects’ infrared spectroscopy, these objects were water-injected beefs and the normal beefs. Furthermore, some characteristic spectrum was collected according to the principle of spectral analysis technology in the wavebands of 900~2200nm, difference of the two classes of object should be emphasized as the reason of the pattern modeling requirements. So wavelet transform was applied to spectral analysis to obtain the singular value which seemed as the chief actor of the difference between water-injected samples and normal ones, and the next step based on singular value was to extract the feature wavebands from spectroscopy of every class of beef. The feature wavebands that contained common group absorption peak were named base value, the others were optional, and different combinations were carried out by them in the end. Clustering methodology provided characteristic spectrum as another additional factor for such combinations above, they were all target matrix which were prepared for pattern recognition model. Above all, training set and testing set were constructed by using leave-one-out cross validation, the optimum displayed identify outcomes of different combinations, and the value of recognition rate was 90.48%. Much analysis about difference significance test was necessary, and then Mann-Whitney method was used to analyze the significance of recognition rate above, it was showed that the result was significant, and the feature wavebands obtained from clustering analysis took pattern recognition model a higher prediction accuracy, in which, wavebands of 1818~1842nm had great influence.