Simple counts of individuals are commonly used in studies of mammalian populations. However, such counts are known to routinely underestimate population size. When used, counts are assumed to be proportional to population size. We tested the validity of that assumption by analyzing capture–recapture data on 5 species of rodents trapped at a single site during monthly sessions from 1973 through 1993. For each 3-day trapping session, we estimated numbers of animals of each species residing on our trapping grid with selected-model and interpolated-jackknife estimators from the program CAPTURE, the modified Lincoln–Petersen estimator, and the Jolly–Seber estimator. We tested for proportionality by fitting a regression line, constrained to pass through the origin, of estimated population size to each of 3 “counts”: numbers of individuals, numbers of captures, and minimum number known alive (MNKA). We then tested validity of these equations by predicting estimated population size from counts from 1994 through 1998. In general, counts were proportional to estimated numbers within a species, so counts were indices of density and yielded similar patterns of population fluctuations. However, regression coefficients, which reflected probabilities of capture, varied among species. Consequently, counts were not appropriate for interspecific comparisons of abundance, even when trapping protocols were invariant. Probabilities of capture also are likely to vary among sites, trapping protocols, and perhaps by seasons, so conditions for counts as valid indices of population size are restrictive.