We investigate the performance of five glacier melt models over a multi-decadal period in order to assess their ability to model future glacier response. The models range from a simple degree-day model, based solely on air temperature, to more-sophisticated models, including the full shortwave radiation balance. In addition to the empirical models, the performance of a physically based energy-balance (EB) model is examined. The melt models are coupled to an accumulation and a surface evolution model and applied in a distributed manner to Rhonegletscher, Switzerland, over the period 1929–2012 at hourly resolution. For calibration, seasonal mass-balance measurements (2006–12) are used. Decadal ice volume changes for six periods in the years 1929–2012 serve for model validation. Over the period 2006–12, there are almost no differences in performance between the models, except for EB, which is less consistent with observations, likely due to lack of meteorological in situ data. However, simulations over the long term (1929–2012) reveal that models which include a separate term for shortwave radiation agree best with the observed ice volume changes, indicating that their melt relationships are robust in time and thus suitable for long-term modelling, in contrast to more empirical approaches that are oversensitive to temperature fluctuations.