This paper focuses on short-term forecasting of solar irradiance. An innovative two-state model is proposed: if the sun is shining, the solar irradiance is estimated with an empirical model fitted on historical data; if the sun is covered, the clear sky solar irradiance is adjusted according to the cloud transmittance. The distinction between these two states is made by the sunshine number, a binary indicator of whether the Sun is covered by clouds or not, previously introduced by Badescu (2002). Sunshine number is the sole quantity effectively forecasted in the model. The general structure of the model and its advantages are discussed. Its performance on real data is demonstrated, and comparison of the model results against classical ARIMA approach applied to clearness index time series, as main competitor, is made. We conclude with an outlook to future developments oriented to improve the model accuracy. (C) 2014 Elsevier Ltd. All rights reserved.