Evaluation of image difference including colour and spatial aspects is often required in colour reproduction industry. The objective of a successful image difference model should have a good agreement with the image difference perceived by observers. The perceived image difference is affected by many factors, such as image characteristics, human vision system, rendering methods and viewing conditions. The current conventional colour difference formulae can not well predict the perceived difference. In this paper, a new experimental method was designed to investigate how image difference is identified by human observers.