In the study long-term variability of various characteristics of forest fires on the territory of na-tional park "Tunkinsky" according to the data of 1974-2016 is analyzed. Statistical properties of the number of fires connected with recurrence and ex-istence of significant autocorrelated communica-tions allowed to offer the technique of modeling of this characteristic with forecasting for 1 year. On the basis of the analysis of long-term dynamics of frequency of the fires, relative fire areas and index-es of inflammability of the wood, and also climatic data the following methods of mathematical model-ing are applied: regression analysis, moving aver-age method, creation of the trend, autocorrelation, autoregression. The moving average method with averaging of 5 years allowed revealing two basis cycles (1975-1985 and 1986-2014) characterizing the number of fires. Two methods of forecasting of the number of forest fires with anticipation are of-fered for 1 year: on the polynomial trend character-izing the recession of the second cycle and the point of a change to a new tendency, and auto-regression model with the previous values for the seven-year period. Weak significant communication of the number of fires with the sums of rainfall for May, by average June temperatures and the sums of temperatures for April - August is revealed. On the basis of the received factorial communications and existence of autocorrelation in the sequence of the number of fires the autoregression model taking into account significant meteorological factors for forecasting of the productive sign is given. The re-sults of modeling by means of these models can be added with predictive estimates of autoregression model taking into account significant meteorological factors to which, first of all, average air temperature of June belongs.