Previous studies indicate that North American gypsy moth,Lymantriadispar (L.), populations are driven by a numerically bimodal replacement rate model: Nt+1/Nt=f(N), wheref(N) is bimodal, resulting in two equilibrium densities. Under this theory, populations are regulated about a low density equilibrium for many years until some perturbation (usually mass immigration) elevates populations to high densities, where they are regulated about a high-density equilibrium until crashing. In this paper, the evidence for and against numerical bimodality in gypsy moth populations is reviewed. The Melrose Highlands data (egg mass densities at 83 plots in New England from 1910 to 1931) were reexamined. These analyses indicated bimodality inf(N) when data were expressed as yearly means of several plots in a zone ≍30 km in diameter, but there was no clear evidence of bimodality in the dynamics at individual plots. Density fluctuations in these relatively small plots (0.07 ha) were instead dominated by apparently random effects. It is hypothesized that short-range dispersal dominates the dynamics of populations at these spatial scales. These results illustrate the importance of spatial scale in the characterization of ecological processes.