Drought is one of the most important abiotic constraints that negatively impacts rice productivity, especially at reproductive stage. Developing drought-tolerant cultivars possessing high yield under a favorable water regime and fairly good yield under drought stress is crucial to sustain farmers' efforts in intensifying rice cropping systems. To develop such cultivars, we launched a marker-assisted recurrent selection (MARS) program, a new approach in rice breeding, where quantitative trait loci (QTLs) for the target traits are mapped within the progeny of elite materials with complementary traits, and favorable alleles of those QTLs are accumulated through successive cycles of intercrossing and genotyping-based selection. A population of 230 F3:5-derived lines from a cross between IR64 and B6144F, and 10 check varieties, have undergone 14 field evaluations for yield under favorable conditions (2 years) during the wet season, and five field evaluations for yield under cyclical drought during the dry season (one year) in five locations (Banfora, Burkina Faso; Longorola, Mali; Ibadan and Badeggi, Nigeria; Los Baños, Philippines and Santa Rosa, Colombia). The population was also evaluated for drought tolerance in a rain-out shelter (ROS), and also genotyped with 484 single nucleotide polymorphism (SNP) markers polymorphic between the two parents. A wide level of variability for target traits, irrespective of the water regime, was observed among the progenies in all locations. Under favorable conditions, the population average yield was 5.06 t/ha in Banfora, 4.19 t/ha in Badeggi, 4.08 t/ha in Ibadan, and 3.9 t/ha under the ROS. Three progenies out-yielded all the checks in Badeggi, and several yielded more than both parents. Similar variability was observed during the trials for drought tolerance. Yield reduction due to drought averaged 72.6%, 72%, 33% and 47%, under the ROS, in Ibadan, Longorola and Banfora, respectively. Stability estimated from environmental varianc