With the rapid development of Internet technology, the size and complexity of the database are continually growing, the traditional classification method can no longer meet the demand of the classification of complex data. For this reason, a data classification algorithm based on variational Bayesian is proposed. This paper introduces the variational approximation theory on the basis of traditional Bayesian inference, combines with the thought of maximum expected algorithm, utilizes the mean field theory in the statistical physics, and simulates taking Gaussian mixture model as an example. The experimental results show that the randomly generated data are composed of the three Gaussian models mixed after 382 iterations, the lower bound of likelihood function rises with the increase of iteration number, the curve becomes flat as expectation after 350 iterations, and the mean value and the inverse of covariance matrix close to the real data are obtained in the range of allowable error. Under the requirement of high precision, the calculation speed is faster, calculation efficiency is higher, and all of these accord with the demands of actual engineering application background.