Jatropha curcas is an interesting alternative for biodiesel production due to the high oil content in its seeds, its ability to grow in a wide range of climate and soil conditions as well as low cost of production. However, the species is considered to be in domestication and there are no defined cultivars. Therefore, it is extremely important to understand the genetic diversity of the species for selection and characterization of promising genotypes to initiate breeding programs. The objective of this study is to evaluate the phenotypic diversity of physic nut in order to select the most divergent and superior genotypes to compose future breeding programs, using multivariate analysis. Eleven agronomic characters were evaluated in 165 J. curcas genotypes from the in vivo germplasm bank, which were: Plant height, stem diameter, number of primary branches, fruit length, width, weight and shape, seed length, width and weight plus the oil content. The data were analyzed by principal component analysis (PCA), cluster analysis by Ward and k-means methods. The character fruit shape was removed from the multivariate analysis as the only one with qualitative character. The PCA resulted in 4 main components (PC), which explained 71.62% of total variance. The characters selected in PC1 were seed weight, fruit width, fruit length and fruit weight. There were 22 promising genotypes highlighted, with potential to be exploited in breeding programs. Cluster analysis by Ward and k-means methods generated 9 groups influenced by all analyzed characters, of which five groups of genotypes had advantageous characters. Regarding fruit shape, 13 genotypes had an ellipsoid lanceolate shape and the others had an ellipsoid spherical shape. Multivariate analyses allowed genotype characterization, indicating good strategies used for the selection in genetic breeding programs.