The formulation of mathematical models for predicting coke quality (M 25 and M 10) is considered. On those principles, existing mathematical models are classified in terms of the parameters that they include (batch, technological, and mixed models) and the data analysis employed (structural, analytical, adaptive, and neural-network models). A similar classification in terms of parameters is generally accepted and intuitively employed. In view of the limited number of measurable technological and batch parameters, classification of models in terms of parameters does not permit the creation of models for M 25 and M 10 that provide qualitatively different assessments of the process or novel insights. The classification of models in terms of the type of data analysis suggests that combining several models based on different approaches may result in a more flexible hybrid model for predicting M 25 and M 10.