The highest amount of genetic variation could be found in local genotypes and also relative wild species. Wild species are known as important resources of favorable genes for improvement of many traits including resistance to biotic and abiotic stresses, oil quality and even yield improvement. Information about genetic control and heritability of traits is one of the prerequisites of progress in plant breeding. Identification of genetic parameters and their various effects such as additive, dominance and their interactions, are very important in determining the breeding strategy. Generation mean analysis is a good method to estimate genetic relationships within and between generations by crossing between pure lines and analyzing different generations. In this study which was done on safflower, one genotype from each of C . tinctorius , C. palaestinus and C. oxyacanthus were selected and crossed for development of F1, F2, F3, BC1:1, BC2:1 and BC1:1S1 generations. To study the differences between and also the effect of drought stress on different generations, composite analysis of variance were done and results showed that there are significant differences among generations for most of the studied traits. Therefore the analysis of these traits were possible through generation mean analysis. After the analysis of different models including one to six parametric models for both crosses and both environments, no significant epistatic effects were detected which might be affected by data transformation, therefore three parametric model were used for all traits and the adequacy of the model was confirmed. Estimate of genetic components of variance showed that less environmental and higher additive variance and as a result higher narrow sense heritability could be detected in this study. F1 heterosis, transgressive segregation and higher drought tolerance than the parental genotypes were detected in the segregated generations especially in t*o cross. Keywords : Safflower, Carthamus species, interspecific cross, generation mean analysis
