Abstract Data on plants from five groups of remnant prairies and forests in the prairie-forest ecotone of the midwestern United States show that: (1) Archipelagos of small sites tend to contain more species than do single large ones of equal total area. (2) No species are excluded from small sites. (3) Small sites tend to have surprisingly many species, and large sites surprisingly few, relative to a random colonisation model. (4) ‘Rare’ species (those that occur in only one site) are found more often in small sites than a random colonisation model would predict. (5) There is no evidence for any of these sites that species number is ‘relaxing’, but all these sites have been remnants for only a short time. Result (1) is consistent with results from similar studies on other taxa and suggests that there is no automatic reason to believe that species richness is maximised by single large refuges.