When it comes to steel processes, it is known that refractory materials are responsible for a significant portion of the steel production costs. For this reason, this work aimed to understand the high variability and the low durability of the refractory campaign that compose a process of continuous casting in a large LD mill in the state of Minas Gerais , identifying the relationship between the process variables so it was possible to make estimates about the duration of its refractory campaign. For the selection of the explanatory factors, a variation of the method Stepwise was used. In each step of the algorithm, a model based on linear programming was responsible for the calculations of the linear regression coefficients. In the end, a prediction model was obtained for the duration of the campaign containing 12 explanatory factors and 97.66% of statistical significance.