Flotation is a very complex process, described by a large number of input and operating variables, frequently high correlated and collinear. Therefore, it is a great challenge to find out which combination of those variables are driving the process out of its target, given by concentrate grade and process recovery. Since most variables are measured with significant error, the on-line identification of which measurement need to be recalibrate is an issue, in order to assure good control decisions. Multivariate statistical projection method, and particularly Principal Component Analysis, can be effectively used to obtain linear models between latent variables directly from operating data. In this work, PCA models have been obtained and used to develop an on-line monitoring system on a pilot rougher circuit. The main experimental results and work to be done are discussed here. (C) 2018, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.