Ozturk, M., Salman, O. and Koc, M. 2011. Artificial neural network model for estimating the soil temperature. Can. J. Soil Sci. 91: 551Á562. Although soil temperature is a critically important agricultural and environmental factor, it is typically monitored with low spatial resolution and, as a result, methods are required to estimate soil temperature at locations remote from monitoring stations. In this study, cost-effective, feed-forward artificial neural network (ANN) models are developed and tested for estimating soil temperature at 5-, 10-, 20-, 50-and 100-cm depths using standard geographical and meteorological data (i.e., altitude, latitude, longitude, month, year, monthly solar radiation, monthly sunshine duration and monthly mean air temperature). These data plus measured monthly mean soil temperature were collected for 2006Á 2008 from 66 monitoring stations distributed throughout Turkey to obtain a total of 2376 data records (36 months)66 monitoring stations) for each of the five soil depths. At each soil depth, 1800 randomly selected data records were used to develop and train a separate ANN model, and the remaining 576 records at each depth were used to test and validate the resulting models. Good agreement was obtained between ANN-estimated soil temperature and measured soil temperature, as evidenced by correlation coefficients of 98.91, 97.99, 99.03, 98.26 and 95.37% for the 5-, 10-, 20-, 50-and 100-cm soil depths, respectively. It was concluded that ANN modeling is a reliable method for predicting monthly mean soil temperature in regions of Turkey where soil temperature monitoring stations are not present.Ozturk, M., Salman, O. et Koc, M. 2011. Un re´seau neuronal artificiel pour estimer la tempe´rature du sol. Can. J. Soil Sci. 91: 551Á562. Malgre´son importance cruciale en tant que facteur agronomique et environnemental, on surveille habituellement la tempe´rature du sol a`une faible re´solution spatiale. Par conse´quent, on a besoin de me´thodes pour estimer la tempe´rature aux endroits e´loigne´s des stations de surveillance. Dans le cadre de cette e´tude, les auteurs ont mis au point puis teste´des mode`les rentables et sans re´troaction fonde´s sur un re´seau neuronal artificiel (RNA) pour estimer la tempe´rature du sol a`5, 10, 20, 50 et 100 cm de profondeur, a`partir des donne´es ge´ographiques et me´te´orologiques usuelles (par ex., altitude, latitude, longitude, mois, anne´e, rayonnement solaire mensuel, dure´e mensuelle de l'ensoleillement et tempe´rature moyenne de l'air mensuelle). Ces donne´es ont e´te´recueillies de 2006 a`2008 avec la tempe´rature moyenne mensuelle du sol dans 66 stations re´parties un peu partout en Turquie, de manie`re a`obtenir 2 376 releve´s (36 mois ) 66 stations de surveillance) pour chacune des cinq profondeurs. À chaque profondeur, on a utilise´800 releve´s, se´lectionne´s au hasard, pour former un mode`le RNA distinct, les 576 autres releve´s servant a`ve´rifier et a`valider les re´sultats. La tempe´rature du sol estime´e graˆce au RNA concorde bien avec ce...