This paper aims to find the best mathematical model that can represent the call holding time which is a statistical indicator in cellular networks, directly affecting network performance and providing critical insight for the network service provider. In this study, a data collection system is designed and empirical voice call data are obtained and analysed. Each user's call holding times are extracted via the proposed system and thus analysis of user behaviors are provided. As a result of our analysis of the obtained distributions, accuracy levels were determined using goodness-of-fit tests and error vector norm based techniques. Consequently, this study confirms that the lognormal family gives the best fitting for the distribution of all and per user's call holding times.