The present study aimed to predict the dry matter digestibility (DMD) of animals fed complex diet using the spectra difference between diets and feces by near infrared reflectance spectroscopy (NIRS). Four indoor experiments were carried out with artificial mixed complex diets in the state field observatory stations located at Hebei province. The sheep were used as tested animals. One hundred and four parts of “diet-feces” were obtained and divided into a calibration sample set (70 samples) and a validation sample set (34 samples). The spectra difference between diets and feces, and the partial least square (PLS) were used to calibrate the models of DMD of sheep. The results indicated that the optimum range of spectrum was 8 656-4 310 cm-1. The optimum proceeding method was multivariate scattering correction + first derivative + Norris smoothing. The optimum number of factors was 7. The root mean square error of prediction (RMSEP) was 2.46%, and the RPD value of validation was above 2.5. The coefficient of correlation between the actual and predicted DMD using the NIRS was 0.8339, indicating that the NIRS could be used to evaluate dry matter digestibility of sheep fed complex diets.