We explored whether a variation in predation and habitat complexity between conspecific populations can drive qualitatively different numerical dynamics in those populations. We considered two disjunct populations of the least killifish, Heterandria formosa, that exhibit long-term differences in density, top fish predator species, and dominant aquatic vegetation. Monthly censuses over a 3-year period found that in the higher density population, changes in H. formosa density exhibited a strong negative autocorrelation structure: increases (decreases) at one census tended to be followed by decreases (increases) at the next one. However, no such correlation was present in the lower density population. Monthly census data also revealed that predators, especially Lepomis sp., were considerably more abundant at the site with lower H. formosa densities. Experimental studies showed that the predation by Lepomis gulosus occurred at a much higher rate than predation by two other fish and two dragonfly species, although L. gulosus and L. punctatus had similar predation rates when the amount of vegetative cover was high. The most effective predator, L. gulosus, did not discriminate among life stages (males, females, and juveniles) of H. formosa. Increased predation rates by L. gulosus could keep H. formosa low in one population, thereby eliminating strong negative density-dependent regulation. In support of this, changes in H. formosa density were positively correlated with changes in vegetative cover for the population with a history of lower density, but not for the population with a history of higher density. Our results are consistent with the hypothesis that the observed differences among natural populations in numerical abundance and dynamics are caused in part by the differences in habitat complexity and the predator community. ?? Springer-Verlag 2005.