Measurements of energy consumption in dwellings show that the consumption is normally distributed. The variation is wide. A statistical simulation method based on the Monte Carlo technique is used to study variation of energy consumption in dwellings. Uncertain parameters such as internal temperature and air change are assumed as distributions. Values for each parameter distribution are combined randomly. The probable energy consumption is simulated and given as a distribution instead as a single value. The method is used to study the influence of variation from climate, building and inhabitants. Variation from inhabitants is much more significant than variation from climate. It is therefore important to know the variation from inhabitants＇ behaviour. If the inhabitants＇ behaviour is unknown, it is impossible to predict the total energy consumption more accurately than +/- 15-20% compared with the consumption found with traditional energy calculation methods. The heating and ventilation energy consumption is encumbered with an uncertainty of +/-25-40% if the inhabitants＇ behaviour is unknown. In a cold climate it is most important to study uncertainties of transmission and ventilation loss.