Late spring frost is a severe risk during early plant development. It may cause important economic damage to grapevine production. In a warming climate, late frost risk either could decline due to the reduction in frost days and an advancement of the last day of frost or increase due to a more pronounced shift forward of the start of the active growing period of the plants. These possibilities were analyzed in a case study for two locations in the lower Swiss Rhone Valley (Sion, Aigle) where viticulture is an important part of agriculture. Twelve phenology models were calibrated for the developmental stage BBCH09 (bud burst) using measured or reconstructed temperature data for two vineyards in Changins (1958 to 2012) and Leytron (1977 to 2014) together with observed phenological data. The day of year (DOY) for BBCH09 was then modelled for the years 1951 to 2050 using the best performing phenology model in combination with ten downscaled and bias-corrected climate scenarios. A 100-day period starting with BBCH09 was defined, during which daily mean and minimum temperatures were used to calculate three frost risk indices in each year. These indices were compared between the periods 1961-1990 (reference) and 2021-2050 (climate change scenario). Based on the average of the ensemble of climate model chains, BBCH09 advanced by 9 (range 7-11) (Aigle) and 7 (range 5-8) (Sion) days between the two time periods, similar to the shift in the last day of frost. The separate results of the different model chains suggest that, in the near future, late spring frost risk may increase or decrease, depending on location and climate change projections. While for the reference, the risk is larger at the warmer site (Sion) compared to that at the cooler site (Aigle), for the period 2021-2050, small shifts in both phenology and occurrence of frost (i.e., days with daily minimum temperature below 0 °C) lead to a small decrease in frost risk at the warmer but an increase at the cooler site. However, considerable uncertainties remain that are mostly related to climate model chains. Consequently, shifts in frost risk remain uncertain for the time period considered and the two study locations.