Drought is a critical constraint for legume production in semi-arid regions, yet breeding for drought tolerance in faba bean through induced mutagenesis remains largely unexplored. To our knowledge, this is the first EMS-derived mutant population in faba bean specifically developed for drought tolerance, comprising 45 M₂/M₃ lines derived from small-seeded cv. Zina and large-seeded cv. Aguadulce Superlonga), evaluated under two irrigation regimes—100% field capacity (well-watered control) and 40% field capacity (severe stress)—over two consecutive growing seasons in a randomized complete block design with three replications. Drought stress caused severe yield losses, reducing mean seed number per plant by 42.2% and mean seed weight per plant by 47.1%. Analysis of variance revealed highly significant effects of genotype, irrigation, and generation/year on both yield components. The non-significant genotype × irrigation interaction indicated similar proportional drought response across genotypes, while the non-significant three-way interaction suggested relatively consistent genotype rankings across generations/growing seasons. Among the ten drought tolerance indices evaluated, seed-number-based mean productivity (MPn) and stress tolerance index (STIn) were the most discriminating, whereas weight-based indices failed to differentiate genotypes due to the inherent seed-size contrast between botanical backgrounds. Dunnett’s comparisons identified genotype 23 (Zina-derived) as the top performer, significantly exceeding its parent for both MPn and STIn; genotypes 22, 24, 12, 3, and 15 similarly outperformed controls. Cluster analysis broadly distinguished three groups: a tolerant cluster dominated by Zina-derived lines, a moderately tolerant cluster (Zina wild-type), and a sensitive cluster of Aguadulce Superlonga-derived lines. These findings suggest that EMS mutagenesis generated potentially heritable and exploitable variation for drought tolerance, with selected lines representing promising candidates for further multi-environment validation.