Abstract. Collection selection is traditionally a sub-problem of metasearch, and identifies collections most likely to contain relevant documents. However, we propose to treat collection selection as an independent search task with the goal of identifying collections that are relevant as a whole; so the user may return to them to serve future (related) information needs. Using a new methodology and framework we evaluate the suitability of existing collection selection algorithms for this search task, compared with a new algorithm designed specifically for the task.