Ecological specialization is widely believed to be a major driver of background extinction rates. One scenario is that species progress through a taxon cycle: as successful forms become more specialized, they decrease in abundance and therefore suffer elevated extinction rates. This model predicts strong correlations between specialization, rarity, and extinction rates. Here I quantify these variables for a set of about 800 Paleocene-Miocene fossil mammal species from the western USA. Only species with age ranges of at least 1.0 m.y. were included. Observed temporal duration was used as a point estimate of extinction rate. Abundance and specialization were estimated using new equations. If N is the species richness of an assemblage falling in the geographic and temporal range of a species, local abundance can be estimated as 1 / N when a species is present and 0 when it is absent; overall abundance is the mean of the local values. Because body mass is the most important ecological attribute in mammals, specialization was estimated as local crowding along the body mass gradient (similar results were obtained using carnassialization or anisodonty). The crowding coefficient C for each species is defined as the sum of exp(-km) across all other species, where kis a constant and m is the difference in log body mass between pairs of species. k is set to minimize the Kolmogorov-Smirnov test statistic D for a comparison between the observed probability density of species as a function of body mass, and an estimated density function that is directly proportional to C. The results are surprising. Large mammals are slightly more, not less, abundant, and there is no monotonic relationship between body mass and duration. Generalized species have slightly lower, not higher, abundance. There is virtually no relationship between either abundance or crowding and duration. By contrast, geographic range size is a fairly strong predictor of duration, explaining > 10% of the variance. The results provide only the weakest support for the taxon cycle hypothesis. Metapopulation dynamics may be more important: perhaps extinction of restricted species results from the stochastic loss of metapopulations, or from geographically localized environmental catastrophes.