In order to uncover the complex relation in the inventory data and lower the inventory level, based on the idea of black box method, all factors that influence demand are identified firstly. Then an initial forecast model of BP neural networks that adopt LM algorithm is established, and is trained using hospital inventory's history data. The interrelation's information of each factor that influences demand is stored in the link weights matrix W dispersedly. The final forecast model is obtained. We use the model to forecast the demand of medical equipment in Daping hospital. Based on it, the inventory cost is reduced enormously. The theory for the inventory system can be used to make management decision.