The persistence of small populations remains a puzzle for ecology and conservation. Especially interesting is how naturally small, isolated populations are able to persist in the face of multiple environmental forces that create fluctuating conditions and should, theory predicts, lead to high probabilities of extirpation. We used a combination of long‐term census data and a five‐year demographic study of a naturally rare, endemic plant, Yermo xanthocephalus, to evaluate the importance of several possible mechanisms for small population persistence: negative density dependence, vital rate buffering, demographic compensation, asynchrony in dynamics among sub‐populations, and source–sink dynamics. These non‐exclusive explanations for population persistence all have been shown to operate in some systems, but have rarely if ever been simultaneously examined for the same population or species. We hypothesized that asynchrony in dynamics and demographic compensation would be more powerful than the other three mechanisms. We found partial support for our hypothesis: density dependence, asynchrony among population segments, and source–sink patterns appear to be the most important mechanisms maintaining population viability in this species. Importantly, these processes all appear to operate strongly at very fine spatial scales for Yermo, allowing the only two, extremely small, populations to persist. We also found considerable differences in the results of our census and demographic analyses. In general, we estimated substantially greater chances of population survival from the census data than from the shorter‐term demographic studies. In part, this difference is due to drier than average climate conditions during the years of the demographic work. These results emphasize that while demographic information is necessary to understand various components of population dynamics, longer term studies, even if much less detailed, can be more powerful in uncovering some mechanisms that may be critical in stabilizing population numbers, especially in variable environments.