The shape of the relationship between population growth rate and population density has important consequences for population dynamics. Past reviews have led to suggestions that, for large mammalian herbivores, the density‐dependent response is curvilinear or abrupt in its onset above some threshold density, and that different demographic segments differ in their sensitivity to rising density. I examined the form of density dependence in stage‐specific mortality rates for three African ungulate populations for which data were available over a sufficiently wide range in abundance: kudu in Kruger National Park, South Africa, and wildebeest in both Kruger and Serengeti National Park, Tanzania. Calf losses remained fairly high toward low‐density levels, indicating that juvenile mortality was sensitive to various influences, some acting independently of density. The maximum population growth rate at low density was truncated by density‐independent factors restricting recruitment success, while the slope of the initial plateau region of the density‐dependent response depended on how steeply juvenile mortality rose and fertility in the youngest reproductive stage fell with increasing density. Adult mortality rose only above some threshold density, especially among older females, and was largely responsible for setting the zero growth level. The density‐dependent response in adult mortality was effectively linear, so that the trend in population growth rate toward the zero growth level was also linear. Differences in the stage of onset and steepness of the density dependence in adult female mortality appeared to be affected by differences in the predator–prey relationship for these three ungulate populations. Overall nonlinearity in the population growth response was an emergent outcome of the different stages at which mortality and recruitment losses among different population segments became affected. Models depicting curvilinear density dependence exaggerate the tendency toward overcompensation around the zero growth level and resultant propensity toward oscillatory dynamics. Reliable projections of population dynamics require models that are faithful to the intrinsic demographic responses determining the overall population growth rate.