In this investigation, video recordings of the underflow discharge of a pilot plant hydrocyclone were collected during classification of different precious metal ores. The underflow shape was monitored by determining the underflow width along a horizontal line through the image. This was accomplished by employing various noise reduction methods and identifying the flow boundaries via motion analysis. Subsequent monitoring of dilute, transitional and dense flow could be automated by embedding the underflow width measurements and making use of one-class support vector machines to estimate the distributional densities of the data in the resultant phase space. Experimental results suggest that the approach could provide a practical and inexpensive means of monitoring the operational states of hydrocyclones. (C) 2011 Elsevier Ltd. All rights reserved.