Clustering, the automated grouping of similar data, can provide powerful insight into large/complex data. We demonstrate the benefits of clustering applied to output from climate model inter-comparison initiatives. We focus on modelled tropospheric ozone from the ACCMIP project. Cluster-based subsampling of the model ensemble can (i) remove outlier data on a grid-cell basis, reducing model–observation bias and (ii) provide a useful framework in which to investigate and visualise model diversity.