We explore two methods that correct for differential sampling effort when estimating the true number of species in a region based on samples such as those typically recorded in museum or conservation databases. The two methods are: (1) a phenomenological model that relies on a saturating sampling curve; and (2) a model based on a lognormal distribution of species abundances. We test these methods using a database for the butterflies of Oregon and find that the distribution of high-diversity areas, using the estimated, or “asymptotic”, diversities, is strikingly different from the geographic pattern one would deduce if the raw data were used, without correcting for differential sampling effort. Further, we show that differences in accuracy exist between the two estimation procedures, and that these differences are aggravated at small sample sizes; we argue that estimates based on the lognormal distribution should be preferred because they can offer substantial improvement over analyses based solely on the raw data, generally without risking overestimation. Lastly, using both the database and estimated values of butterfly diversity, we show that the distribution of endangered and numerically rare butterflies rarely coincides with “hotspots” or centers of biodiversity. Thus, protecting regions of Oregon rich in overall butterfly diversity will not normally protect the bulk of rare or endangered butterfly species. Keywords: asymptotic diversity, biodiversity analyses, butterflies, conservation databases, sampling effort.