AFLP technique was used to assess genetic diversity in 72 soybean varieties under cultivation in India. Selected 12 AFLP primer pairs produced 1319 products of which 1257 were polymorphic (95%). Wide variations were observed for the number of amplification products, percent polymorphism and average polymorphism information content (PIC). The Jaccard's similarity indices (J) based on the AFLP profiles of the 72 soybean varieties were subjected to UPGMA cluster analysis. The dendrogram generated revealed four major clusters, which were strongly supported by the high bootstrap values obtained from analyses of 1000 bootstrap samples. In addition, the Mantel's test for cophenetic correlation with r = 0.955 indicated very good fit of the varieties to a group in the cluster analysis. Some correspondence between the clustering pattern and the pedigree, place of release or target area of the variety was observed. Overall moderately high genetic diversity was observed which appears to be due to the higher genetic diversity prevalent in 12 of the varieties included in three diverse clusters and was indicative of the need to include more diverse germplasm in the soybean improvement programs.