This study was conducted to investigate the relationship among the agronomic traits in some safflower genotypes (20 genotypes) and to evaluate the efficiency of selection indices based on seed yield components. The experiment was carried out as a randomized complete block design with three replications at the research farm of Isfahan University of Technology in 2010. The results showed that there were significant differences among the genotypes for all of the studied traits, except for days to 50% emergence. Also high broad-sense heritabilities and low differences between genetic and phenotypic coefficients of variation were found for all of the traits except days to 50% emergence. Genetic and phenotypic correlation coefficients showed that seed yield components of capitula/plant, number of seeds/capitulum and 1000-seed weight highly and positively correlated with seed yield. The results of stepwise regression for seed yied/plot showed that number of capitula/plant, number of seeds/capitulum and 1000-seed weight explained 87 percent of its variation. For seed yield/plant, 1000-seed weight, number of capitula/plant and number of seeds/capitulum were important respectively, and explained 70 percent of the variation for this trait. The results of path analysis for seed yield/plot and seed yield/plant showed that number of capitula/plant, number of seeds/capitulum and 1000-seed weight had high direct effects on seed yield. Cluster analysis classified the genotypes into three distinct groups of 10, 5 and 5 genotypes. The highest expected response to direct selection was estimated for number of capitula/plant, thus direct selection for this trait can be more effective than the other yield components. Relative efficiency of indirect selection for seed yield/plot via seed yield/plant and number of capitula/plant was high, indicating that indirect selection via these traits can be effective to improve seed yield/plot. Relative efficiency of indirect selection for seed yield/plant via number of capitula/plant, number of seeds/capitulum and 1000-seed weight was also high, suggesting that these traits can be used as selection criteria to improve seed yield/plant. In this study, the selection indices of Smith-Hazel 1, Smith-Hazel 2 and Pesek-Baker were determined based on the number of capitula/plant, number of seeds/capitulum and 1000-seed weight. High genetic correlation of these indices with seed yield/plot and seed yield/plant suggested that they can be effectively used to improve seed yield. The expected gain from selection for these selection indices were almost the same, indicating that they can be used to simultaneously improve number of capitula/plant, number of seeds/capitulum and 1000-seed weight.