The analysis of the building performance in representative locations is generally exploited to get useful insights and data supporting the drafting of energy policies for the existing building stock. Regardless of the criteria adopted for climate classification, only typical weather conditions are usually considered, with a loss of information on climate variability and on its impact on building energy performance. The lack of extremes, moreover, not only affects the outcome of the climate classification and the set of locations identified as representative but also prevents a robust analysis of the buildings’ resilience and, thus, of their ability to adapt to climate change. To investigate this aspect, extreme years have been considered in this research as alternative to the typical ones to perform a weather-based classification and select representative locations. This study was based on a multi-year (2008–2022) ERA5-Land dataset with 480 municipalities. After generating typical and extreme reference years, a hierarchical clustering has been implemented, using monthly weather statistics of temperature, humidity, and solar irradiance as inputs. After selecting a set of representative climates for both typical and extreme cases, EnergyPlus simulations have been run for Brazilian residential building archetypes, registering energy needs for space heating and cooling and operative temperature distributions. The results showed that extreme weather files determined clusters with higher air temperatures and lower humidity levels. Regarding the simulation outcomes, the operative temperature results showed that extreme weather files increase the operative temperature, leading to a reduction the heating energy needs but increasing the ones for cooling.