Flotation processes are very complex, and after more than one hundred years of history, there are few reports on applications of novel techniques in monitoring and control of flotation units, circuits and global plants. On the other hand, the successful application of multivariate predictive control on other processes is well known. In this paper, an analysis on how the characteristics of flotation processes, the quality of measurements of key variables, and the general lack of realistic dynamic models, are delaying the appropriate use of predictive control. In this context, the applications of multivariate statistics, such as PCA, to model the relationship between operating data for on-line diagnosis and fault detection and to build causal models are discussed. Also the use of PLS models to predict target variables for control purposes, is presented. Results, obtained at pilot and industrial scales, are discussed, introducing new ideas on how to obtain more valuable information from the usual available operating data of the plant, and particularly from froth images. © 2010 Elsevier Ltd. All rights reserved.