Abstract: Abstract: The production of raw milk is a dynamic and continuous process. There can be some new changes in the quality of produced raw milk every day. A delay determination can occur in the product safety system when assessing veterinary drug residues in the subsequent procedure using the current detection data. The purpose of this study aims to construct an early warning system based on the full-scale historical data, which can be used to identify, control and evaluate veterinary drug residues in raw milk, and then make accurate judgments before risks occur. Three parts include in this system. 1) The raw milk samples from 50 pasture in Shanghai were first taken as the population to be sampled based on statistical principles. The population was divided into different parts in each pasture. The simple random sampling method means to take samples from various parts, and combine them to obtain the final screening sample. Specifically, the fresh milk samples from 50 pasture in Shanghai were taken once a week, and a total of 300 screening samples were obtained for 6 weeks in May and June. The screening results of veterinary drug residues in these raw milk samples indicated the contamination conditions of veterinary drugs during the sampling period. 2) The collected samples were then tested. The detection test was performed on the high performance liquid chromatography, high resolution time-of-flight tandem mass spectrometry(HPLC-TOF-MS/MS) by mixing standard working solutions. A total of steroidal veterinary drugs (12 types), glucocorticoids (6 types), sedatives and quinolones (5 types each) were detected in this study. There were 42 types of veterinary drug residues in total associated with 12 categories of veterinary drugs. Six veterinary drugs were detected in every batch: Thiabendazole, Altrenogest, Beclomethasone, Carazolol, Methyltesterone, and Norethandrolon, indicating the contamination of the veterinary drug in these pastures. 3) A risk warning system was finally constructed. When the number of collected samples was fixed for the screening of the veterinary drug residues in raw milk for each batch from the pastures, the number of prednisone detected（Pn）control chart can be chosen to establish a risk early warning system based on the results of screening of veterinary drug residues. During the 6-week screening, the Pn control chart showed a stable state. When the number of detected prednisone was 6 in the 7th week, it infers that the Pn control chart still showed a stable state. When the prednisone detection number was 10 in the 7th week, the Pn control chart was abnormal in the unsteady state, indicating the batch of samples triggered the warning. All the results show that the established risk warning system can be real-time monitor. This proposed system can play an early warning and preventive role in the occurrence of safety risks for the veterinary drug residues in raw milk.