Abstract: An issue of interest in forecasting electricity demand is to understand how it reacts to changes in explanatory variables like income, number of households and relative prices. Data for them are only available on a monthly basis, and monthly models arise as the natural tool for estimating the relevant elasticities and forecasting the load for horizons, say, from one month to two years ahead. The demand of electricity, however, is also dependent on the weather conditions and the number and type of special days, weekends, holidays and vacation periods, in a month. The standard practice to account for their contribution is to include monthly aggregates of weather and special day’s indicators in the right hand side of the equation. No attention is usually paid to the nonlinear and switching-regime effects of those variables on daily demand and on the income and price elasticities. We advocate for a two-step procedure that combines a daily model and a monthly model, to take into account the different speed of reaction of the demand to the changes in its determinants. In the first step a daily model is constructed to derive efficient estimates of the effects of weather conditions and special days, and to compute a daily series of demand adjusted for those effects (AD) and series of demand due to temperature effects (TD) and to special days (SD). Agrregate these three demands at monthly level. Secondly, monthly models for AD and SD are specified, income and price elasticities are estimated, and medium-term forecasts for them are obtained. Forecast TD as a function of temperature. Final forecasts for the observed load are derived by summing the forecasts of the different demands. We illustrate our proposal by applying it to the demand of electricity in mainland Spain.