Durum wheat is an important crop for the human diet and it is consumed largely in the form of traditional dishes such as pasta, couscous, unraised breads, and bourghul. Worldwide, the consumption of durum wheat is gaining popularity because of its nutritional properties. In order to ensure that durum wheat production maintains the pace with the increase in demand, it is necessary an increase in production of approximately 1.5% per year. Genomic selection (GS) recently appeared as an ideal tool to deliver this level of annual genetic gain, but its deployment in durum wheat breeding remains extremely limited. Here, we present a pilot study aimed at refining the mode of potential deployment of GS in durum wheat breeding. Four RILs populations were used to train statistical model incorporating relationship matrix, the genotype by environment interaction, and marker by environment interaction. A total of 576 individuals were sown at three locations in Morocco and Lebanon for collection of yield and yield components. The same individuals where genotyped by sequencing to obtain 3202 polymorphic markers. Our results confirm that the prediction accuracy depend on heritability of the trait, the size of the populations, the proportion of TP and VP and relatedness between TP and VP. Also, feeding the model with information on markers associated with significant QTLs increased the overall accuracy.