Owing to their complexity, many metallurgical processes are difficult to model fundamentally, and as a result more attention has been focused on the use of artificial intelligence techniques to simulate these processes. In this paper recent advances in the use of neural nets to model processes not amenable to conventional methods are discussed. These include the application of a new modelling methodology based on the use of neural nets for the dynamic simulation of metallurgical processes, the use of conventional techniques in conjunction with neural nets to simulate gravity separation circuits, as well as the detection of gross errors in variable measurements by means of a neural net representation of the relation between measurement and constraint residuals.