Although its theoretical foundation is well understood by researchers, a blast furnace is like a black box in practice because its behavior is not always as expected. It is a complex reactor where multiple reactions and multiple phases are involved, and the operation heavily relies on the operators' experience. In order to help the operators gain insights into the operation, the authors do not use traditional metallurgy models but instead use machine learning methods to analyze the data associated with the operation performance of a blast furnace. They analyze the variables that are connected to the economic and technical performance indices by combining domain knowledge and results obtained from two fundamental feature selection methods, and they propose a classification algorithm to train classifiers for the prediction of the operation performance. The findings could assist the operators in reviewing as well as improving the guideline for the operation.