Modeling of user behaviour becomes almost a necessity on data networks nowadays. Besides using their resources more efficiently, many data communication systems will perform more effective via estimation of statistical distribution of packet counts transferred (uploaded or downloaded) by users at once. Hence, the purpose of this study is to search suitable statistical models for distribution of packet counts that is transferred by user at once, on a arbitrary network. For this aim, an Android application is used to track Internet usage of real users. A detailed statistical analysis is performed using this data and the results are checked by applying different performance tests. Consequently, it is observed that the most suitable statistical distribution to model user behaviour is lognormal family.