A method of predicting the operation reliability of a distribution network, the method being based on an ARIMA model. By establishing an ARIMA model predict a user monthly power outage count; convert a non-smooth element outage count time series to a smooth time series; perform regression on only lagged values of a dependent variable, and on current values and lagged values of a stochastic error term, so as to establish a user monthly power outage count model; according to a prediction result sample an outage point, and at the same time, taking into account a real time load operating state, establish a failure mode effects table based on TLOC criteria and PLOC criteria; and calculate a system recovery time for each instance of device outage, and finally obtain a whole-year reliability index.