Ecologists commonly use matrix models to study the population dynamics of plants. Most studies of plant demography use plot-based methods to collect data, in part, because mapped individuals are easier to relocate in subsequent surveys and survey methods can be standardized among sites. However, there is tremendous variation among studies, both in terms of plot arrangement and the total area sampled. In addition, there has been little discussion of how alternative sampling arrangements influence estimates of population growth rates (lambda) calculated with matrix models. We surveyed the literature to determine what sampling designs are most used in studies of plant demography using matrix models. We then used simulations of three common sampling techniques-using a single randomly placed plot, multiple randomly placed plots, and systematically distributed plots-to evaluate how these alternative strategies influenced the precision of estimates of lambda. These simulations were based on long-term demographic data collected on 13 populations of the Amazonian understory herb Heliconia acuminate (Heliconiaceae). We found that the method used to collect data did not affect the bias or precision of estimates in our system-a surprising result, since the advantage in efficiency that is gained from systematic sampling is a well-known result from sampling theory. Because the statistical advantage of systematic sampling is most evident when there is spatial structure in demographic vital rates, we attribute this result to the lack of spatially structured vital rates in our focal populations. Given the likelihood of spatial autocorrelation in most ecological systems, we advocate sampling with a systematic grid of plots in each study site, as well as that researchers ensure that enough area is sampled-both within and across sites-to encompass the range of spatial variation in plant survival, growth, and reproduction.