Accurate digital simulation is crucial in food packaging design, helping designers predict the performance and appearance of packaging. Traditional digital simulation methods often require a large number of calculations and manual adjustment of parameters, which is not only inefficient but also prone to errors. The neural networks were used in this article to establish a simulation model for food packaging so as to address this issue. This model learned the physical and appearance characteristics of packaging, and was trained and optimized through back propagation algorithms. A large amount of food packaging dataset was used for training in the experiment, and the simulation results of traditional methods and neural network methods were compared. The results showed that simulation methods based on neural networks could more accurately predict the performance and appearance of food packaging, and had higher efficiency and stability, reaching 92.7% and 94.6% respectively. This has provided a new method and tool for food packaging design, which is expected to be widely applied in practical applications.