Although cone classifiers have received little attention in the modelling of mineral processing units they are one of the oldest non-mechanical classifiers. Their use has enjoyed a revival in recent years because of their low cost of installation and operation. In this paper the development of models for cone classifiers is reviewed and a new model is proposed. The experimental work, data collection and analysis are briefly described, and the new model is presented and discussed. The data used in the model development are made up from three distinct experiments using different units in each case. There are 21 data sets from a 3 m cone treating the tailings in a copper concentrator, 24 data sets from a 0.3 m pilot-scale cone, and 29 data sets from a laboratory glass funnel (0.(Y73 m diameter) used as a cone classifier, giving a total of 74 data sets. The model structure was developed from an analysis of particle settling in classifiers, using specialised model building software. The size of a particle settling in a separator and the forced flow through the apex of the cone were related via hydrodynamic dimensionless groups to the operating conditions of the unit. The model predicts the separation efficiency using the following four parameters: efficiency curve shape factor, cut-point, underflow density (per cent solids) or underflow pulp velocity. The model covers a range of feed flowrates from 1.2 to 5000 1/min.