In order to achieve on-site, rapid and relatively accurate acquire the total nitrogen (TN) content of manure in the large-scale dairy farm that is going through the whole sections from the barn to the farmland before, 111 manure samples during the whole process (collection-separation-stacking) were collected from a typical large-scale dairy farm in Tianjin lasting for 6 consecutive days. All samples were dried using electrothermal blast dryer, rushed and sifted through 18 meshes. The total nitrogen contents were measured by Kjeldahl nitrogen analyzer, and the range of concentration is 0.20%~3.86%. Near-infrared (NIR) diffuse reflectance spectra of all samples were collected in the range of 4 000~12 000 cm^-1 using the PerkinElmer Fourier NIR spectrometer. 17 outlier samples were removed based on Monte Carlo cross validation method. The NIR diffuse reflectance spectra of the remaining 94 samples were pretreated by SG first derivative and denoising. Then, the principal component analysis was adopted to obtain the variation of samples in the whole process of fecal treatment in large-scale dairy farm. The first two principal components (PCA) can explain 89% of the total variance. The results of PCA showed that the properties and components of fresh feces of dairy cows in different reproductive ages were similar. From feces to manure, the properties and components of fresh feces were not changed any more. However, the properties and components were greatly changed in the bedding stage. Therefore, it is necessary to establish a global TN quantitative analysis model suitable for the whole process of fecal treatment for realizing the real-time and rapid detection of TN in the whole process of fecal treatment in large-scale dairy farms. 94 samples were divided into calibration and prediction sets based on K-S method. 63 samples, including 24 fresh manure samples, 28 mixed manure samples and 11 cushion samples, were used as calibration set to construct a calibration set. The partial least squares model for quantitative analysis of TN was established in the whole process of fecal treatment in large-scale dairy farms. 31 unknown samples, including 12 fresh manure samples, 9 mixed manure samples and 10 cushion samples, were predicted by the established global model. The correlation coefficient (R) between the predicted concentration and its actual concentration was 0.91, and the root means square error of prediction (RMSEP) was 0.151%. The research showed that it was fully feasible to determine the TN contents in the manure of large-scale dairy farms integrated the near infrared diffuse reflectance spectroscopy with chemometrics, not only providing the theoretical and experimental basis for the development and field application of a near-infrared instrument for rapid detection of TN in the manure, but also the support for quantitatively recycling the manure to the farmland.