In bread wheat, genome wide association study was conducted for four traits (days to anthesis, grain filling duration, grain number/ear and grain weight/ear) relevant to drought tolerance. For this purpose, a spring wheat reference set (320 genotypes) was genotyped for ~18K SNPs using genotyping-by-sequencing (GBS), and phenotyped under irrigated (IR) and rain-fed (RF) environments. Four different models that were used for GWAS included the following: (i) single locus single trait (SLST) mixed linear model (MLM); (ii) SLST statistical machine learning (SML) approach, (iii) multi-locus mixed model (MLMM), and (iv) multi-trait mixed model (MTMM). A variable number of MTAs were identified for the four traits using two SLST approaches (543), MLMM (702) and MTMM (559); 15 MTAs were common among MTAs identified using different approaches. Together, only 57 MTAs (12 from SLST; 24 from MLMM and 21 MTMM) were stable in RF and IR+RF environments. Further, 42 significant epistatic interactions were identified for the three traits (DTA, GFD and GWPE) under different environmental conditions. Out of 42 interactions, 37 were identified under RF and five were identified under IR conditions. For individual trait, the number of epistatic interactions ranged from 2 for days anthesis to 20 for grain weight per ear. The MTAs identified during the present study supplemented the available information on the genetic control of traits under IR and RF conditions. Stable MTAs may be exploited for improvement of drought tolerance in wheat using MAS. Seven candidate genes associated with stable MTAs and involved each in one of the following processes were also identified: transcription, transferase activity, protein processing, positive regulation of catalytic activity, lipid metabolic process, protein serine/threonine kinase activity and cytochrome P450 oxidase activity. Further studies involving these candidate genes for drought responsive traits may prove useful.