Crop models are frequently used in ecology, agronomy and environmental sciences for simulating crop and environmental variables at a discrete time step. The aim of this work was to test the predictive capacity of the Delphi system, calibrated and determined for each pedoclimatic factor affecting durum wheat during phenological development, at regional scale. We present an innovative system capable of predicting spatial yield variation and temporal yield fluctuation in long-term analysis, that are the main purposes of regional crop simulation study. The Delphi system was applied to simulate growth and yield of durum wheat in the major Italian supply basins (Basilicata, Capitanata, Marche, Tuscany). The model was validated and evaluated for three years (1995–1997) at 11 experimental fields and then used in operational mode for eleven years (1999–2009), showing an excellent/good accuracy in predicting grain yield even before maturity for a wide range of growing conditions in the Mediterranean climate, governed by different annual weather patterns. The results were evaluated on the basis of regression and normalized root mean squared error with known crop yield statistics at regional level.