Solid concentrations, calculated from slurry density measurements, particularly in hydrocyclones streams, are important to both grinding circuit control and flotation control. Due to the multiplicity of grinding lines, slurry density sensors in each line would be too expensive and would require additional maintenance efforts. For this reason it is common practice in the grinding plants not to have real sensors in the overflow and underflow streams but only in the feed streams. With the aim of making possible the online availability of solid concentration in overflow streams, the design of a soft-sensor in an actual grinding plant is addressed and presented in this paper. To develop such a soft-sensor, an ARMAX model structure is developed using the stepwise regression method. The method selects, one by one, and in a systematic manner the variables which best model the measurement to be replaced choosing them from a list of presumably correlated candidate variables. In this paper not only the single measurements are included in the list of candidates, but also combinations of such measurements having physical significance. A phenomenological model has been built with the purpose of determining these combinations. Actual grinding plant data is used to determine the model structure, to estimate the model parameters and to test the predictive capability of the developed soft-sensor.