Running a successful reagent plant trial is not an easy task, especially when trying to achieve an incremental improvement in a highly variable process, such as flotation. Cytec aims to achieve one per cent recovery improvement with 90 per cent statistical confidence. Most concentrators experience large fluctuations in recovery and so the magical' one per cent is difficult to detect. With the aid of statistical software, including REFDISTR (proprietary) and ECHIPR, Cytec have developed a unique approach to differentiate between daily operational variation and actual process improvement. This paper describes how Cytec runs a plant trial based on extensive laboratory investigation and detailed analysis of historical plant data using statistical techniques.