The main objective of the present study was to compare three se­lection methods: (1) conventional method based on mean yield alone (YA), (2) Kang's rank-sum (KRS) method, and (3) Yield-stability statistic (YSi) method. The other objective was to study the nature of stability. It was based on genotype-environment (GE) interaction partitioned to het­erogeneity due to environmental index (additive main effect) and residu­al. The analysis of variance of data set made up of 28 genotype eval­uated over six environments indicated significant genotype x environ­ment (GE) interaction for seed cotton yield, boll weight and earliness. While, it was insignificant for lint yield. The removal of heterogeneity re­vealed that genotype (H111414/96) was unstable for seed cotton yield, three genotype (H5 1204/96, H8 1379/96 and G.87) were unsta­ble for boll weight and genotype (H6 1304/96) was unstable for earli­ness because of the linear effect of environment index. For certain stable genotype, linear and non-linear effect of the environmental index were significant. The KRS method identified the highest number of stable gen­otypes. The mean seed cotton yield of selected genotypes were 7.92 kentar/feddan for YA method. 7.74 kentar/feddan for KRS method and 7.87 kentar/feddan for yield stability statistic. The fifty bolls weight were 129.10, 126.93 and 126.01gm for the three selection methods, respectively. The mean earliness were 76%, 73.30% and 74.034% for these selection methods, respectively. These reductions were regarded as insignificant considering the breeder would be able to choose more consistently performing genotypes on the basis of KRS than on the basis YA or yield-stability statistic method. Selection of highest genotypes of yield-stability statistic improved this method.