Many organizations collect vast quantities of three-dimensional (3-D) scientific data in volumetric form for a range of purposes, including resource exploration, market forecasting, and process modelling. Traditionally, these data have been interpreted by human experts with only minimal software assistance. However, such manual interpretation is a painstakingly slow and tedious process. Moreover, since interpretation involves subjective judgements and each interpreter has different scientific knowledge and experience, formulation of an effective interpretation often requires the cooperation of numerous such experts. Hence, there is a pressing need for a software system in which individual interpretations can be generated automatically and then refined through the use of cooperative reasoning and information sharing. To this end, a prototype system, SurfaceMapper, has been developed in which a community of cooperating software agents automatically locate and display interpretations in a volume of 3-D scientific data. The challenges and experiences in designing and building such a system are discussed. Particular emphasis is given to the agents’ interactions and an empirical evaluation of the effectiveness of different cooperation strategies is presented.