Shelf life prediction models of cookie were established using dynamic method, weibull hazard analysis method and BP artificial neural network method respectively. Comparison results of the three models showed that the prediction result of BP artificial neural network model is more accurate and convenience; the correlation coefficient of BP model, R2, between the outputs and targets is 0.963 8; BP neural network can incorporate various factors, including food product compositions, package properties, storage and logistics condition into a single model.