Breeding for water stress tolerance is very difficult as the stress conditions created in experimental set up and actual field conditions, invariably, have poor correspondence. Direct selection for yield under stress holds promise but is slow on account of complexity of yield. Screening based on yield suffers from low heritability. Yield has been the primary breeding objective in production breeding and has been improved either by targeting yield per se as well as yield components based on correlated response. Drought stress indices are quantitative measures that characterize water stress response by yield data from one or several environments based on timing, duration and intensity of stress. Such an index is more readily useable than raw yield data. Since drought resistance is a yield based trait, selection could vary depending on which index is chosen by the breeder. A major drawback of using these indices has been lack of correspondence in rankings across indices and their failure to discriminate overlapping responses in terms of yield under stress. There has been substantial effort on part of scientists to develop and validated different indices based on yield per se, index scores, regression analysis as well as various model based approaches. The major focus has been on identifying variables such as phenology, physiological traits, biomass partitioning or other parameters that influence crop performance under stress. The paper appraises the approaches from yield based indices to regression based as well as crop model based that have been used to understand crop response to water stress.