The Cayeli volcanogenic massive sulphide deposit is located in Black Sea region of Turkey. Cayeli Bakir Isletmeleri (CBI) was formed as a joint venture between Inmet (Canada) and Eti Holding (Turkey). CBI is processing complex copper/zinc sulphide ores. The Cayeli concentrator was commissioned in 1994, with a design capacity of 600 000 tonnes/year. The grinding circuit was a classical rod mill/ball mill combination. The rod mill was converted to a primary ball mill in May 1997. The reasons for conversion were to use mill power more efficiently and eliminate feed interruptions during rod charging. The grinding circuit now has a primary ball mill in open circuit, and a secondary ball mill in closed circuit with hydrocyclones. Classification was originally done in two stages with hydrocyclones. Both mill discharges were combined and pumped to a primary cyclopack having two, 500 mm diameter cyclones. Primary cyclone overflow was pumped to a secondary cyclopack for further classification. The secondary cyclopack had ten, 200 mm cyclones. Both unit underflows were fed to a secondary ball mill by gravity. The secondary cyclone overflow, typically 62-64 per cent passing 36 m, at around 42 per cent solids (w/w) was fed to flotation circuit. In order to increase circuit capacity to 1 000 000 tonnes/year and to improve control over the grinding circuit, it was decided to convert the classification circuit from two stage to one stage. For this purpose, six Warman Cavex cyclones having 400 mm diameter were installed. Now, the combined primary and secondary ball mill discharge was classified in single stage with the Warman cyclones in closed circuit with the secondary ball mill. The cyclone overflow goes directly to the copper rougher circuit as flotation feed. The grinding circuit is producing a flotation feed of 70 per cent passing 36 m. The change over to single stage classification at Cayeli was successful. A mill throughput of 1 000 000 tonne/year is being achieved with better control over classification, and improved grind size. Results and comparison between the two classification circuits are presented in the paper.