The work aims to select the optimal model from various models and achieve parameter identification. The package model was constructed as the parameter uncertainty model. The model parameters were identified with the Markov Chain Monte Carlo method in Bayesian inference framework. Deviation information criterion (DIC) was used to calculate the DIC parameters of each alternative model and select the optimal package model. The mass block-buffer material was used to simulate the package and the random vibration test was carried out on the vibration test bench. The analysis result showed that the Bouc-Wen model (n=2) was the optimal model for the package. The model selection and parameter identification method proposed based on Bayesian inference takes the uncertainty of the model into account. The constructed model can accurately predict the time-domain signal of acceleration response of the package under random vibration.