Stable carbon isotope composition (δ13C) of dry matter has been widely investigated as a selection tool in cereal breeding programmes. However, reports on the possibilities of using stable oxygen isotope composition (δ18O) as a yield predictor are very scarce and only in the absence of water stress. Indeed, it remains to be tested whether changes in phenology and stomatal conductance in response to water stress overrule the use of either δ13C or δ18O when water is limited. To answer this question, a set of 24 genotypes of bread wheat (Triticum aestivum) were assayed in two trials with different levels of deficit irrigation and a third trial under rainfed conditions in a Mediterranean climate (northwest Syria). Grain yield (GY) and phenology (duration from planting to anthesis and from anthesis to maturity) were recorded, and the δ13C and δ18O of grains were analysed to assess their suitability as GY predictors. Both δ13C and δ18O showed higher broad-sense heritabilities (H2) than GY. Genotype means of GY across trials were negatively correlated with δ13C, as previously reported, but not with δ18O. Both isotopes were correlated with grain filling duration, whereas δ18O was also strongly affected by crop duration from planting to anthesis. We concluded that δ18O of grains is not a proper physiological trait to breed for suboptimal water conditions, as its variability is almost entirely determined by crop phenology. In contrast, δ13C of grains, despite being also affected by phenology, still provides complementary information associated with GY.