Data reconciliation is extensively applied to improve the accuracy and reliability of plant measurements. It relies on process models ranging from simple mass and energy conservation equations to complete causal models. The precision of reconciled data mainly depends on the complexity and quality of plant models used to develop data reconciliation observers. In practice, the difficulty of obtaining detailed models prevents the application of powerful observers like the Kalman filter. The objective of this study is to propose a methodology to build a model for a flotation circuit to support the implementation of a Kalman filter for dynamic data reconciliation. This modeling approach extracts essential information from the plant topology, nominal operating conditions, and historical data. Simulation results illustrate that applying a Kalman filter based on a rough empirical model that has been correctly tuned gives better estimates than those obtained with sub-model based observers. (C) 2015 Elsevier Ltd. All rights reserved.