Complementarity‐based algorithms for the selection of reserve networks emphasize the need to represent biodiversity features efficiently, but this may not be sufficient to maintain those features in the long term. Here, we use data from the Common Birds Census in Britain as an exemplar data set to determine guidelines for the selection of reserve networks which are more robust to temporal turnover in features. The extinction patterns found over the 1981–1991 interval suggest that two such guidelines are to represent species in the best sites where they occur (higher local abundance) and to give priority to the rarer species. We tested five reserve selection strategies, one which finds the minimum representation set and others which incorporate the first or both guidelines proposed. Strategies were tested in terms of their efficiency (inversely related to the total area selected) and effectiveness (inversely related to the percentage of species lost) using data on eight pairs of ten‐year intervals. The minimum set strategy was always the most efficient, but suffered higher species loss than the others, suggesting that there is a trade‐off between efficiency and effectiveness. A desirable compromise can be achieved by embedding the concerns about the long‐term maintenance of the biodiversity features of interest in the complementarity‐based algorithms.