In human science, we often face problem of estimating human behavior parameters. Behavior frequency is one of the most used behavior indicators. Knowing the behavior frequency, we can draw conclusions about significant aspects of behavior both in the present and in the future. It is often impossible to obtain data about behavior frequency directly. Therefore, there is a problem to estimate behavior frequency on limited data such as data about some last episodes of behavior. We propose two models based on Bayesian belief networks to estimate behavior frequency. We use last three behavior episodes, maximum and minimum intervals between the episodes as initial data and test the models on the data set collected from the social network Vk.com.