Three recent studies (Kisi, 2006a(Kisi, ,b, 2007a) ) intended to model reference evapotranspiration (ET 0 ) by neural network systems. The studies were published in Nordic Hydrology (NH), Hydrological Sciences Journal (HSJ) and Hydrological Processes (HP), subsequently. These studies will be denoted as the NH, HSJ and HP papers, respectively. The NH paper uses the feed-forward back propagation (FFBP) artificial neural network (ANN) with Levenberg-Marquardt (LM) and conjugate gradient (CG) training algorithms together with multiple linear regression (MLR) technique and applies these models to the climatic data, wind speed (W), solar radiation (SR), air temperature (AT ), soil temperature (ST ), and relative humidity (RH ) taken from the Blythe Northeast Weather station in the Californian CIMIS database. The HSJ paper uses the generalized regression neural network (GRNN) method together with the CIMIS Penman, Ritchie and Hargreaves empirical ET 0 models and employs the same climatic data, excluding ST, from the Pomona and Santa Monica weather stations, both located in California. The HP paper also uses the FFBP ANN with LM training algorithm together with the Penman, Hargreaves and Turc empirical ET 0 models and applies the models to the same climatic data (W, SR, AT and RH ) from the Claremont, Pomona and Santa Monica weather stations in California. While this comment was under review, a fourth paper (Kisi, 2008) (to be denoted as HP2) has become available online on 6 November 2007 in Hydrological Processes. Results in HSJ have completely been merged with the HP2 paper. When both papers are compared, it is seen that Figures 2, 3 and 4 of the HSJ paper have been inserted into Figures 5, 6 and 7, respectively, in HP2.