This study explores the applicability of data-driven clustering analysis in predicting vegetation distribution over two continents where water is an important controlling factor for vegetation growth, South America and Africa, and compares the ability of clustering analysis with that of a physically based dynamic vegetation model to predict vegetation distribution. A clustering analysis algorithm based on the genetic-algorithm-based K-means is tested, with the number of clusters determined a priori according to the primary plant functional types observed to exist in the study domain. The most important variables upon which the clustering analysis is based include available water, its seasonality, and evaporative demand. The dynamic vegetation model used is the Community Land Model version 3 coupled with a Dynamic Global Vegetation Model (CLM3-DGVM) with modifications targeted to address some known biases of the model. Results from both the clustering analysis and the modified CLM3-DGVM are compared against observations derived from the Moderate Resolution Imaging Spectroradiometer (MODIS). Both methods reasonably reproduced the general pattern of dominant plant functional type distribution. There is no clear winner between the two methods, as the DGVM outperforms the clustering analysis approach in some aspects and is outperformed in others. It is therefore suggested that clustering analysis can be a useful tool in biogeography estimation, although it cannot be used in mechanistic studies as the process-based DGVMs are.