This paper presents a comparative study of three different modeling techniques for predicting cement fineness using input-output SCADA measurements of the closed circuit grinding in a cement plant. The modeling approaches used are the following: statistical, artificial neural networks (ANN), and adaptive neuro-fuzzy inference system (ANFIS). The data set for generating the predictive models are obtained from a database of the operation of the cement plant, UCEM-Guapan located in Azogues, Ecuador. Online validations of the proposed models allow the selection of the best approach and the most accurate models for cement fineness prediction, Blaine and percentage passing the sieve No. 325.