The probabilistic distribution of cascading outages is one of the main measures to describe the propagation of cascading outages, and to evaluate the risk of the large scale outages of the power system. Historical outage data has always been used for power system reliability evaluation, and by combined with the branching process model, it is used for cascading outage analysis for a regional power grid. Based on the 14-year utility historical outage data from a regional power grid in China, several known probabilistic models are tested and compared, and a Borel-Tanner branching process model is proposed to estimate the probabilities of cascading line outages. Statistical error analysis is performed to study the effectiveness of applying the Borel-Tanner model to practical grid risk management. Results indicate that the empirical distribution of the total number of line outages is approximated well by the Borel-Tanner model. For the same confidential level, the estimation of the probability distribution of the larger cascades by the Borel-Tanner branching process model requires significantly fewer recorded outage data than empirical estimation by a factor of 10-1.