There are few clearly described utility studies in advanced ovarian cancer, despite the public health importance of condition and the need for preference based measures of quality of life in economic evaluation of the new treatments. We used data clustering techniques to develop health state descriptions based on data from 66 women who completed the EORTC QLQ-C30 over a six month period while receiving chemotherapy for ovarian cancer. The health state descriptions were presented to a group of members of the general public (n=38), via the internet, and preferences elicited using the standard gamble technique. Mean utility values ranged from 0.685 to 0.977, although the range of individual preferences was wider, including values as low as 0.125. This is the first study to use data clustering methods combined with internet preference elicitation in oncology. The resulting health state model is parsimonious, data driven, and incorporates quality of life items tailored to cancer. The estimates therefore meet the needs of policy makers while reflecting more accurately the experience of disease than those based on generic preference measures.