We describe a procedure to build an artificial neural network model of half-hourly values of diffuse solar radiation at the surface that can be repeated for other locations in a region. The model was developed for the location of the Portoroz Airport (Slovenia) using data gathered by a standard automatic meteorological station and diffuse solar radiation measured over one year. The model was constructed based on a perceptron artificial neural network, which is a universal approximator for highly nonlinear systems. To date, models of this type have been restricted to a single chosen location. An inland location at Maribor was tested as a benchmark for comparison. It is shown that the Portoroz model can be directly transferable without significant quality loss to the inland location of Maribor Airport, which has a different climate. Comparison to the Maribor benchmark model gives a correlation ranging from the initial value of 0.9030 to 0.9004, RMSE increases from 40.5 to 43.7 Wm(-2), coefficient of variation of the RMSE increases from 38% to 41%; values for the initial location of Portoroz are 0.9453, 28.7 Wm(-2), 26.4%. To the best of our knowledge, this report describes the first such regional model that is spatially transferable. (C) 2016 Elsevier Ltd. All rights reserved.