Abstract: Aiming at the hysteresis and destructiveness of off-line static detection of critical quality attribute of the moisture content of the raw material unit of the traditional Chinese medicine manufacturing process, honey-processed Tussilago farfara, honey-processed Astragalus and honey-processed Glycyrrhiza uralensis were used as the research carriers, and the drying method was used to measure the moisture content as a reference value. The moving stage was used to simulate the movement process of samples on the conveyor belt in the actual on-site production process, and near-infrared (NIR) spectra were collected, combined with machine learning, to establish NIR on-site dynamic detection model of moisture content in multi-variety honey-processed Chinese herbal slice. The results show that the second derivative method is used to preprocess the spectrum. The number of decision trees (ntree), the number of random features (max feature), and the minimum number of samples for generating leaf nodes (node size) are selected: 46, 76, and 8, respectively. The quantitative analysis model of moisture content has the best effect. The prediction coefficient of determination (the prediction coefficient of determination, R_pre²) and the root mean square error of prediction (root mean square error of prediction, RMSEP) of the model were 0.903 2 and 0.330 2, respectively. The NIR quantitative model for the moisture content of multi-variety honey-processed Chinese herbal slice established in this study has good predictive performance, and can achieve rapid, accurate and non-destructive quantitative analysis of the moisture content of honey-processed Tussilago farfara, honey-processed Astragalus and honey-processed Glycyrrhiza uralensis at the same time, and provides a method for determining the moisture content of honey-processed Chinese herbal slice of the raw material unit of the traditional Chinese medicine manufacturing process.