Issues related to power quality of electrical service are becoming more and more concerning, not only because of the impact that they have both for the industrial production system and for the service industry in a globalized and competitive world, but also because of the regulatory authorities' penalty systems that are pushing Distribution System Operators (DSO) to ensure better service standards for customers. That being so, it is not surprising that DSO are trying to better understand the reliability performance of their electrical distribution networks; most of the interruptions to which the customers are subjected originate from medium voltage power lines, which exhibit a different annual count of outages depending on various factors, one of them being the unpredictable weather impact. In this work an outage annual frequency model is developed in order to present a reliability analysis of medium voltage power lines; this model is effective in pointing out the reliability differences between different sub-networks and is suitable for a chronological comparison of performances. For this reason, two type of Negative Binomial regressions are considered in order to exploit the high number of data collected from feeders outage records, always keeping the distinction between long and short duration outages.