Frost damages develop when exposure overtakes frost vulnerability. Frost risk assessment therefore needs dynamic simulation of frost hardiness using temperature and photoperiod in interaction with developmental stage. Two models, including or not the effect of photoperiod were calibrated using five years of frost hardiness monitoring (2007–2012), in two locations (low and high elevation) for three walnut genotypes with contrasted phenology and maximum hardiness (Juglans regia cv Franquette, Juglans regia x nigra ‘Early’ and ‘Late’). The Photothermal model predicted more accurate values for all genotypes (Efficiency = 0.879; RMSEP = 2.55 °C) than the Thermal model (Efficiency = 0.801; RMSEP = 3.24 °C). Predicted frost damages were strongly correlated to minimum temperature of the freezing events (ρ = −0.983) rather than actual frost hardiness (ρ = −0.515), or ratio of phenological stage completion (ρ = 0.336). Higher frost risks are consequently predicted during winter, at high elevation, whereas spring is only risky at low elevation in early genotypes exhibiting faster dehardening rate. However, early frost damages, although of lower value, may negatively affect fruit production the subsequent year (R2 = 0.381, P = 0.057). These results highlight the interacting pattern between frost exposure and vulnerability at different scales and the necessity of intra-organ studies to understand the time course of frost vulnerability in flower buds along the winter.< Réduire