An understanding of the level of genetic diversity is a prerequisite for designing efficient breeding programs. Fifty-one cultivars of four cotton species (Gossypium hirsutum, G. barbadense, G. herbaceum and G. arboreum) representing core collections at four major cotton research stations with a wide range of eco-geographical regions in India were examined for the level of genetic diversity, distinct subpopulations and the level of linkage disequilibrium (LD) using 1100 amplified fragment length polymorphism (AFLP) markers with 16 primer pairs combinations. The AFLP markers enabled a reliable assessment of inter- and intra-specific genetic variability with a heterogeneous genetic structure. Higher genetic diversity was noticed in G. herbaceum, followed by G. arboreum. The genetic diversity in tetraploid cotton species was found to be less than that in the diploid species. The genotypes VAGAD, RAHS14, IPS187, 221 557, Jayhellar of G. herbaceum and 551, DLSA17, 221 566 of G. arboreum were identified as the most diverse parents, useful for quantitative trait loci (QTL) analysis in diploid cotton. Similarly, LRA 5166, AS3 and MCU5 of G. hirsutum and B1, B3, Suvin of G. barbadense were most diverse to develop mapping populations for fibre quality. The internal transcribed spacer sequences were sufficient to resolve different species and subspecies of diploid cotton. Low level of genome-wide LD was detected in the entire collection (r2 = 0.07) as well as within the four species (r2 = 0.11–0.15). A strong agreement was noticed between the clusters constructed on the basis of morphological and genotyping data.