This study compares geostatistical interpolation and stochastic simulation approaches for the estimation of daily global solar radiation (GSR) on a horizontal surface in order to fill in missing values and to extend short record length of a meteorological station. A guideline to select an approach is suggested based on this comparison. Three geostatistical interpolation models are developed using the nearest neighbor (NN), inverse distance weighted (IDW), and ordinary kriging (OK) schemes. Three stochastic simulation models are also developed using the artificial neural network (ANN) method with daily temperature (ANN(T)), relative humidity (ANN(H)), and both (ANN(TH)) variables as predictors. The six models are compared at 13 meteorological stations located across southern Quebec, Canada. The three geostatistical interpolation models yield better performances at stations located in a high density area of GSR measuring stations compared to the three stochastic simulation models. The guideline suggests an optimal approach by comparing a threshold distance, estimated according to a performance criteria of a stochastic simulation model, to the distance between a target and its nearest neighboring station. Additionally, the spatial correlation strength of daily GSRs and the at-site correlation strength between daily GSRs and the predictor variables should be considered. (C) 2016 Elsevier Ltd. All rights reserved.