To realize the rapid prediction of total volatile basic nitrogen （TVB-N） content in mutton， near-infrared spectroscopy technology and computer vision technology were employed to obtain the spectral information and image characteristic parameters， and least squares support vector machine （LSSVM） prediction model for TVB-N content was established based on fused spectra and image information. Spectral information in the range of 320-1 100 nm and image information of 73 mutton samples storage at 4 ℃ from 1 to 15 days were collected， and the TVB-N contents in samples were determined referring to the national standard method. Then the spectral characteristic information was extracted using competitive adaptive reweighted sampling algorithm， and the color and texture features were extracted as the image characteristic information. Finally， spectral information and image information were fused using feature layer fusion method， and LSSVM models were established based on three kinds of characteristics information， respectively. The results showed that the model based on fused spectra and image information yielded better prediction performance than that based on spectral information or image information merely， with correlation coefficient in the prediction set of 0.930， the standard analysis error of 1.873 mg/100 g， and relative percent deviation of 2.635. The results indicated that the feasibility of the prediction for TVB-N content in mutton based on spectra and image information. It provided an effective method for quantitative， rapid， nondestructive and accurate prediction of TVB-N content in mutton samples.