Abstract: Phenological records obtained for wheat (Triticum durum Desf., cv Karim ) sown at 26 different dates around the year in no limiting water and nutrient conditions were used to investigate the predictive ability of the accumulation heat models and their sensitivity to thermal amplitude. Monitoring of days of Emergence (E), Tiller initiation (T), Jointing (J), Heading (H) and Maturity (M) and meteorological data concerned two different thermal regimes sites: Tunis (18 series) and Kef (8 series). Results show the performance of seven phenology models: the Number of calendar Days (ND), Growing Degree-Days (GDD), Modified Growing Degree-Days (MGDD), Photothermal Units (PTU), Solar Radiation (SR), Solar Thermal Units (STU) and Simplified Beta Function (SFB). For the seeding-emergence and emergence-tiller initiation stages, the GDD gave the best result with significantly the same average index value for the two stations. Because the rate of development is a linear function of temperature, there is no effect of thermal amplitudes on model performances. However, in the case of all phenological phases following tiller initiation all the tested models are sensitive to thermal amplitude presenting significantly different average indices values for the two stations which bring as to conclude that the temperature-growth relationships are non linear. The adjustment of the heat unit models with mathematical functions of thermal amplitudes improves their predictive accuracy. The adjusted thermal indices give significantly the same average values with less variability.