Drought stress is considered as one of the main important factors reducing field crop yield, thus leading researchers to study the related physiological characteristics may reduce drought effects. This experiment was conducted in research farm of seed and plant improvement institute (SPII) in Karadj, Arak and Jolgeh Rokh for three cropping seasons (2008-2011) to evaluate drought tolerance of bread wheat genotypes through irrigation cut-off from anthesis and 20 days after anthesis comparing to full irrigation in 19 winters and facultative wheat genotypes using RCBD in split plot arrangement with 3 replicates. The main and subplots were irrigation regimes and wheat genotypes, respectively. The statistical analyses of coefficients of correlation, stepwise regression, path analysis, factor analysis via principal component analysis were done to discard insignificant or less effective and to consider the most important characteristics directly or indirectly influencing grain yield. Path analysis was only done on independent variables suggested by the stepwise regression model. The final study of the relationships among the variables was according to the results of principal component analysis and varimax rotation. The results showed significant correlations of spike and biomass weight (0.86**), biomass weight with seed number per spike (0.89**) and also seeds number per spike with seed weight per spike in normal condition. The number of seeds per spike had significantly positive correlation with spike and biomass weight characteristics (0.92**). In S2 condition the highest correlation attributed to grain yield with seed number per spike (0.79**). In this condition, there was a significant and positive correlation between seed number per spike and seed weight per spike (0.88**). In S1 condition significant and positive correlations of grain yield with TKW (0.76**) and harvest index (0.78**) and also with seed number per spike (0.69**) were observed. The highest direct effect of seed number per spike was seen in all three conditions and stepwise regression also showed the same result. Further path analysis results showed that yield components of seed number per spike, seeds weight per spike and TKW incorporated in the path analysis model as the potentially effective characteristics in wheat grain yield improvement program in normal and terminal drought conditions. Due to direct and indirect effects of seed number per spike on grain yield, it can be considered as an effective characteristic in the wheat breeding program. Principal component analysis results confirm that boosting 1^st factor related characteristics of yield and yield components have the largest effect on grain yield and should be considered as the best selection criteria for improvement.