This study aims towards an improved estimation of annual heat demand of the building stock for an entire region. This requires the holistic representation of aspects influencing the heat demand of buildings, namely their geometry, fabric, users and surrounding environment. A large data base for the building stock of the Swiss canton of Geneva was systematically assessed to identify parameters suited for representation of these aspects. Due to the expectable differences in heat demand, the building stock was categorized into 8 building types. For each type a multiple linear regression model was developed to predict the heat demand. An aspect which has so far been neglected by regression models of buildings’ heat demand is the influence of microclimate. Since this aspect is considerably influenced by the surrounding topography, parameters suited for the representation of the urban topography were defined and included in the regression. The regression analysis revealed that all models were able to explain high shares of the variance (R²: 71.2% to 88.9%). The mean average errors for hotel, health-care, educational and office buildings were ranging between 30.2% and 39.8% while the error for residential buildings was 17.8%. The suitability and of the selected parameters for heat demand prediction was analyzed in detail for the residential building model and revealed that almost all chosen parameters were highly suited.