This paper explores some of the problems encountered in using a data archive to build tools for training radiologists to interpret breast screening images. We detail our experiences of taking images and case notes created as part of the work of breast cancer screening and using them as resources for training. Four instances of the use of the archive in training are described in detail and the problems they reveal are discussed. We formulate some general lessons for the mobility and re-use of rich ensembles of data and artefacts drawn from complex professional settings. We argue for a richer representation of the context from which the data was taken than can be achieved through making selected relations explicit in metadata. We also conclude that facilities for correcting and elaborating data should be available at the point of use, and not separated out as distinct activities.