I examine the suggestion of Holling (1992) that distributions of body sizes for animals show clumps corresponding to basic biotic and abiotic processes operating at different scales in time and space. This is done by comparing the properties of observed sample distributions with what is obtained from computer—generated samples taken from distributions with different numbers of modes. A combination of kernel density estimation and smoothed bootstrap resampling provides a test for whether a distribution with k + 1 modes fits significantly better than a distribution with k modes. The analysis of three data sets discussed by Holling shows sample size distributions that are consistent with the hypothesis that the underlying distributions are unimodal or bimodal. A fourth data set has a sample distribution that seems unusual even for observations from a distribution with eight modes.