In this paper the identification and control of a rougher flotation process is studied using an adaptable hybrid-neural model. The model is based on first principles and a PCA neural network is used for flotation kinetics estimation. Initially, the hybrid model is used for the identification, from input/output data obtained with a realistic phenomenological model, of a series of four flotation cells. Then, different regulatory and optimizing multivariable control alternatives are developed and tested on the process. The control problem is adaptively solved as an optimization problem, using predictions for the steady state obtained using the hybrid model. Results obtained for different input perturbations, setpoint changes and optimization tests show satisfactory performance, satisfying all required objectives without off-set or oscillation. Based on these results, the hybrid model can be considered an excellent option for the identification and control of flotation plants, from the point of view of flexibility and robustness.