Durum wheat is a strategic crop for food security in semi-arid regions such as North Africa, where climate variability increasingly threatens agricultural productivity. Multi-trait selection represents a valuable approach to developing superior genotypes capable of maintaining high performance under such challenging conditions. This study aimed to compare the efficiency of traditional and modern multi-trait selection indices in identifying elite durum wheat genotypes. A total of 59 genotypes were evaluated under field conditions in Sétif, Algeria. Ten traits related to plant growth and agronomic performance were assessed, and a 15% selection intensity was applied. The classical Smith–Hazel (SH) selection index was implemented under two scenarios—retaining multicollinearity (SH_1) and removing multicollinearity (SH_2)—and compared with two modern ideotype-based indices: the Factor Analysis and Ideotype Design-Based BLUP (FAI-BLUP) and the Multi-Trait Genotype Ideotype Distance Index (MGIDI). Among the four indices, the MGIDI and FAI-BLUP achieved the highest total predicted selection gains (36.28%), substantially outperforming SH_2 (30.76%) and especially SH_1 (–35.98%), which was negatively affected by multicollinearity. Modern indices promoted substantial gains in key productivity-related traits, including biological yield (5.64%), grain yield (5.49%), and straw yield (15.1%). Notably, genotypes G42, G10, G26, and G4 were consistently selected across the MGIDI, FAI-BLUP, and SH_2, confirming their superior multi-trait performance and breeding potential. In contrast, SH_1 yielded inconsistent and negative results, highlighting the limitations of applying traditional indices without accounting for multicollinearity. These findings confirm the robustness, efficiency, and practical value of ideotype-based indices—particularly the MGIDI and FAI-BLUP—for multi-trait selection in durum wheat breeding programs, especially under semi-arid environmental conditions.