We present a new approach to clustering based on the observation that ``it is easier to criticize than to construct.'' Our approach of {\em semi-supervised clustering} allows a user to iteratively provide feedback to a clustering algorithm. The feedback is incorporated in the form of constraints which the clustering algorithm attempts to satisfy on future iterations. These constraints allow the user to guide the clusterer towards clusterings of the data that the user finds more useful. We demonstrate semi-supervised clustering with a system that learns to cluster news stories from a Reuters data set. %This paper presents semi-supervised clustering, a new approach to %clustering that allows users to provide advice to the clustering %algorithm that guides it towards clusterings they prefer. %Semi-supervised clustering begins with traditional, fully unsupervised %clustering to find an initial clustering of the data. The clustered %data is then presented to the user so that they may critique it. User %feedback provides a set of constraints that the system tries to %satisfy to find a new clustering that the user prefers. This process %of presenting clustered data to the user, and refining the clustering %in response to user feedback, is repeated until the user is happy with %the clusters. We present a clustering algorithm that learns from user %feedback to find a clustering metric that yields clusters the user is %happy with. We demonstrate the algorithm on the Reuters and 20 %Newsgroups domains.