Electric power consumption is known at fine temporal scales (e.g. hourly) for geographical zones corresponding to the electric network service divisions (e.g. substations). In several applications, there is a strong need to estimate the past or to forecast future consumption at different divisions, for example the town, district or city block levels. The deployment of smart meters only gives a partial answer to this problem because they usually do not provide exhaustive measures at such temporal scales. We propose in this paper a generic approach to estimate electric consumption on any geographical zones from source zones where fine-grained consumption data is available, using in addition socio-demographic information. The approach is evaluated on both real and simulated data.