Aim. Industrial safety (OS) is the state of protection of operating personnel from harmful effects of manufacturing processes, energy, equipment, objects, conditions and schedule of work [1]. The most efficient evaluation of OS in railway transportation is ensured by composite indicators, one of which is the risk assessment indicator. That is also reflected in the Russian legislation that stipulates the requirement to evaluate fire, occupational and other types of risks that affect industrial safety. According to the definition set forth in GOST 33433-2015 [2] risk is a combination of the probability and consequences of an event. The most complicated task related to risk assessment is the choice of the evaluation model for the probability of an undesired event. The model must enable practical applicability of evaluation results for planning of risk compensation measures. Currently, there are a large number of probability evaluation methods that can be divided into two large groups, i.e. expert and quantitative. Expert methods have several well-known shortcomings. The quantitative methods require the construction of a system of equations or an analytical model. In the context of railway facilities the construction of analytical models of probability evaluation is of principal interest due to the possibility of demonstration of the factors that are taken into consideration by the model. The aim of the article is to formalize the analytical method for evaluation of the probability of railway facility transfer into a hazardous state (in the context of industrial safety). Methods . Undesirable events that cause industrial safety incidents in railway facilities are random; they can be represented as a random process. A random system development process, including objects transition from a safe state into hazardous (undesirable) states, i.e. system state change in time, can under some assumptions be described with a semi-Markov process. In general, the construction and solution of semi-Markov models comes down to building a system of homogenous differential equations. This procedure always involves mathematical difficulties. [3] shows the possibility of representation and solution of semi-Markov models with a coupled graph model. Such models are highly visual, and allow formalizing the wanted system states, as well as paths of transition from safe into hazardous states. The main problem of modelling random processes of industrial safety state changes is the requirement to identify the complete list of hazardous states and preceding non-hazardous or pre-hazardous states. The processes typical to railway facilities are characterized by a multitude of states that cause various events. The concept of “state” usually characterizes an instantaneous image, a “cross-section” of a system. Thus, at the first stage of construction and solution of a model of random process of a system’s industrial safety state change, the finite sets of safe and hazardous states of the railway facility under consideration are identified in accordance with the known hazardous state criterion [4]. As the process of state change of a system’s industrial safety in railway transportation is random in time, in this article system operation is described with a semi-Markov process with the assumption that the discrete process is described with an embedded Markov chain. The set of system states and their connections are represented with a directed state graph with defined topological concepts [3]. For a constructed model, the article provides the proof of the theorem identifying the probability of system transition from an initial non-hazardous into a hazardous state, as well as the formula for calculation of such probability. Results . The graph method for evaluation of industrial safety in railway facilities developed in this article includes both the rules of construction of a system’s safety states graph and the tool for evaluation of the probability of system transition into a specific hazardous state. The graph is the basis of the practical method for calculation and forecasting of industrial safety incidents. The article provides the proof of the theorem identifying the probability of system transition from an initial non-hazardous into a hazardous state, as well as an example of application of graph method for evaluation of probability of fire in a fixed facility. The proposed probability evaluation method can be used in planning of industrial safety measures in terms of specification of new states or rules of transition into associated states.