Data logged on mineral beneficiation processes during normal operation are analysed using time series (correlation and spectral) methods to identify the dynamics of the processes for on-line computer control. Recursive least squares estimation techniques are used to identify parametric models for a cassiterite flotation circuit and a coal washery. Identification problems such as the determination of model order, process time delays and the 'contamination' of the data by process recycle streams (feedback) are discussed. Typical results are presented in which optimised models are identified from data logged during normal operation.