In recent years great efforts have been made in the optimisation of ball mill operation. This paper describes a methodology developed for modelling and optimising the steady-state grinding process. Thirteen basic similarity numbers of the process are derived. Based on these similarity numbers, we build a group of mathematical models which give a full description of the relation between dependent and independent variables. The parameter estimation and multi-objective optimisation of the grinding process have been done using these models, and validated with experiments in a continuous laboratory-scale ball mill.