Conventionally, building energy performance is evaluated through energy simulations using a single input weather file referring to present weather conditions. However, the analysis shown in this study demonstrates the high sensitivity of calculated energy consumption to weather files chosen for simulation. Thus, we propose that multiple present and future weather files must be incorporated as random instances of an unknown population, i.e. the climate. This paper describes a methodology to assess the robustness of different energy efficient refurbishments over possible climate projections, by taking into account uncertainties in weather files. The innovation consists in the discussion of energy outcomes in terms of ranges instead of single values.