Abstract: The article represents the methodological basis and justification of crisis assessment at coke enterprises based on a preventive approach, which is aimed at dividing the enterprises of coke industry into classes of the crisis on the basis of the minimum set of diagnostic indicators. Subtractive method of cluster analysis is used for reducing the diagnostic indicators system to the minimum necessary set. The purpose is to carry out express-assessment of the crisis situation at the enterprises. Express-assessment was made by the method of Ward and k-means and allowed to distribute all coking enterprises into the three classes of crisis: potential, medium and deep crisis. In order to determine the probability of transition to classes of crisis for the investigated companies in the next period the forecasting of diagnostic indicators was made, it was supported by using of short-term adaptive forecasting methods. Recognition of the predicted enterprises’ situation was carried out by discriminant analysis and neural networks through the comparing the quality of the recognition between both models. Neural networks appeared more appropriate tool for this task, so method was used to detect the predicted class of crisis for the coke enterprises. Adequacy testing for all of the models showed its high level. Proposed complex of models allowed quickly identify crisis stage of the enterprises that meets the requirements of preventive management.