Fossil floras are an important source of quantitative terrestrial paleoclimate data. Many paleoclimate estimates are based on relationships observed in modern vegetation between leaf morphology and climate, such as the increase in the percentage of entire-margined species with increasing temperature and the increase in leaf size with increasing precipitation. An important question is whether these observed relationships are universal or regional; for example, recent studies suggest that significant differences exist between floras from three domains: the Northern Hemisphere, New Zealand/Australia, and subalpine zones. Also, debate exists over which statistical models of modern data sets, univariate or multivariate, provide the most accurate estimates of paleoclimate. In this study, 12 foliage samples from living Bolivian forests are compared with data sets from different regions. Models based on data sets from North America and Japan, namely the Climate-Leaf Analysis Multivariate Program (CLAMP) data set of J. A. Wolfe, and from east Asia produce reasonably accurate estimates of temperature and precipitation, suggesting that the climate–leaf morphology relationships for Bolivian vegetation do not differ significantly from those for Northern Hemisphere vegetation. The mean leaf size for a given mean annual precipitation is smaller than for a data set from the Western Hemisphere and Africa, but this difference is most likely due to different sampling methods. As for estimating climate from fossil floras, these results, along with the analysis of four other regional data sets, imply that the most accurate climate estimates will be produced by the predictor data set with the most similar climate–leaf morphology relationships. Unfortunately, our present lack of understanding of why climate-morphology relationships vary between the North America/Japan, New Zealand/Australia, and subalpine domains makes it difficult to identify data sets similar to paleofloras. Until we learn more, it is probably best to compare fossil floras to predictor data sets from the same domain. The performance of the various statistical methods depends on the nature of the predictor data set. Multiple regression analysis tends to produce the most accurate estimates for small data sets with a narrow range of environmental variation that have similar relationships to the flora, and linear regression or canonical correspondence analysis for the larger and more varied CLAMP data set. If a similar predictor data set is not available, then nearest-neighbor analysis can still produce accurate paleoclimate estimates.