Individuals of Eriogonum abertianum Torr. (Polygonaceae) flower in spring, or following onset of summer rains, or both. Within populations flowering time is mainly environmentally determined: there is little genetic variance for flowering time, and experimental moisture limitation significantly delays flowering. In the field a Sonoran Desert population experienced significantly more mortality during the foresummer droughts, and had a significantly greater proportion of spring-flowering plants, than a Chihuahuan Desert population. Greenhouse experiments suggest a genetic basis for differences in size and time of flowering between these populations. Fossil and biogeographic evidence support an adaptive interpretation of earlier flowering in the Sonoran Desert. A model of selection comparing spring-plus-summer flowering with spring-only flowering suggests that expected summer fecundity may not offset the risk of foresummer mortality in the Sonoran population. Rather than switching to a spring-only habit as predicted by the model, the species' range ends where summer rainfall declines abruptly. The invariance of the spring-plus-summer habit is not explained by the demographic, historical, or genetic data. Plants which live for more than a year in the wild have offspring which, in the greenhouse, live longer than the offspring of the general population. This suggests a genetic basis for the occasional observed perennation. Analysis of a quantitative genetic model suggests that when adult survivorship is low, selection will generally reduce perennation. The annual habit is thus likely to persist even in the presence of genetic variation for perennation. Optimal control models of plant carbon allocation are extended to include within-season mortality and allometric growth constraints. When parameters are varied in numerical experiments, resulting predictions for easily measurable characters (e.g., time to first flower) often vary only slightly; most differences are in fitness, suggesting that satisfactory empirical tests may be difficult to conduct. Arbitrary mortality functions can optimally lead to multiple flowering episodes, and this can depend sensitively on parameter values. Optimal trajectories with allometric constraints are divided into a period of vegetative growth and another period of mixed growth.