AbstractDurum wheat (Triticum durum Desf.) is a fundamental staple food for the countries of the Mediterranean basin. Climate change is predicted to cause a trend of increasing drought severity in this region in the near future, necessitating the improvement of durum wheat's resilience to drought stress. Using polyethylene glycol to simulate water scarcity, early vigour parameters in germinating seeds are quickly, easily and affordably assessed. Many screenings, however, only consider the seedling stage; consequently, genotypes identified as promising for cultivation in drought scenarios, may not show such features if drought appears in later phenological phases, as happens in Mediterranean climatic areas, generally prone to terminal drought. The correlation between drought stress resistance during the seedling stage (early vigour) and later stages in the life cycle is elusive due to the lack of scientific efforts. Here we used polyethylene glycol screening to classify fifty‐five tetraploid wheat accessions into three clusters (susceptible, medium resistant and highly resistant to drought), based on morpho‐physiological traits. These accessions included durum wheat cultivars and landraces, as well as ancestors like durum emmer wheat and wild emmer wheat. The results of the screenings were combined with subsequent pot experiments using nine randomly selected accessions, imposing terminal drought, and evaluating their performance. Principal component analysis was performed on data for net photosynthesis, stomatal conductance, transpiration and grain yield. Notably, the genotypes that performed best in the pot experiments were also those that performed well in the screening. Highly resistant candidates had in fact higher physiological and performance parameters than susceptible candidates. In summary, polyethylene glycol screening of germinating seeds resulted to be suitable to predictively evaluate drought resistance in tetraploid wheat accessions under terminal drought conditions, typical of Mediterranean climate zones. The reported data, thus evidence of how this inexpensive and simple method might be efficiently applied for large‐scale phenotyping.