With the progress of industrialization of the food processing industry, related production control software has been rapidly applied in this industry, but the master production scheduling (MPS) software of other industries cannot meet the characteristics of the MPS for food processing industry. In view of the above problems, a two-level MPS method based on Q learning is proposed, which adopts a multi-level method and is divided into bloc level, company level and workshop level. According to the order requirements and related constraints, the customer order is converted into the corresponding weekly production schedule and issued to the company level. The company level adopts the state and action of the Q learning algorithm as the basis, combined with the evenly distributed remainder method to make a choice. Then according to the reward function, the trained Q value is updated to the Q table, and the scheduling is selected. The scheduling selection is formulated into a master plan, and the compiled master plan is issued to the workshop level. After receiving the plan, the workshop level conducts scheduling according to the plan. The simulation test using sample data of food processing enterprises verified that the utilization of multi-level, Q learning algorithm and evenly distributed remainder method can solve the problem of MPS in the food processing industry.