Multi-spectral images of Phaseolus vulgaris leaves at the wavelength range of 400~720 nm with an interval of 5 nm are captured by using a multi-spectral imaging system which mainly consists of liquid crystal tunable filter (LCTF) and CMOS camera. Firstly，according to the principles of image brightness and band index, the value of band index and identifiability for Phaseolus vulgaris leaves are calculated respectively among every band. Then, through sorting the value of band index and identifiability for Phaseolus vulgaris leaves, it can be concluded that bands 545, 630, 645, 720, 650 and 570 nm have preferable identification with considering the characteristics of discrete gray levels and rich brightness of images and little correlation coefficient among different bands. Finally, the classification accuracy for Phaseolus vulgaris leaves is calculated according to the principles of minimum Euclidean distance and minimum spectral angle matching. The classification accuracy of characteristic bands for Phaseolus vulgaris leaves is 100.00% and 83.33% separately through using these two methods. We can draw a conclusion that these bands have ideal classification accuracy. Therefore, bands 545, 630, 645, 720, 650 and 570 nm can be used as feature bands for Phaseolus vulgaris leaves.