A method for operational forecasting of fires is proposed that enables the sequential implementation of five procedures. The method development is necessary to predict early fires in premises in order to take measures to prevent them from escalating into an uncontrolled combustion phase ‒ a fire. As a result of research, it was found that a short-term forecast of the recurrence of increments of the air conditions by one step, based on the current measure of recurrence, is an effective indicator of early fires in premises. At the same time, it was found that before the moment of ignition of the material, the state of the air environment is characterized by dynamic stability, which is described by an irregular and time-dependent random change in the recurrence of the states of the vector of current increments of the state of the air environment. The values of the indicated levels of recurrence of the state increments are determined by the probability levels of 0.67 and 0.1, respectively. The probability of recurrence of state increments of 0.67 is characteristic of a larger number of measured states. When the material is ignited, the dynamics of the probability of recurrence of state increments change abruptly. There is a transition from two to one level of recurrence, close to zero probability ‒ the loss of dynamic stability (in the region of count 250). Further dynamics are characterized by the appearance of separate random recurrent increments corresponding to the instability of the air environment in the premises. In the course of the experiment, it was found that the accuracy of predicting a fire by the proposed method ranges from 4.48 % to 12.79 %, which generally indicates its efficiency. The obtained data prove useful in the development of new systems that early warn of fire in premises, as well as in the modernization of existing systems and means of fire protection of premises