The objective of this study was to investigate the relevance of using the in situ volumetric water content at field capacity () as a predictor of the water retention properties by comparing the performances of pedotransfer functions (PTFs) established using artificial neural networks (ANN-PTFs) and support vector machines (SVM-PTFs) with much simpler PTFs in the form of simple linear regressions (SLR-PTFs). A dataset comprising 456 horizons collected in soils located in France was used. The available data were: the silt and clay contents (SC), the organic carbon content (OC), the bulk density at field capacity (), the in situ gravimetric water content at field capacity) related to by using , and the volumetric water content at-1,-3.3,-10,-33,-100,-330 and-1500 kPa matric potential. The performances of the PTFs studied were compared by using the root mean squared error (RMSE) and the coefficient of determination (R²). Our results showed the relevance of using , which was proved to be close to the volumetric water content at-10 kPa matric potential, as a predictor of the water retention properties. With ANN-PTFs, the best performances were recorded when both and SC were used as input data (RMSE = 0.027 cm 3 cm-3 and R 2 = 0.92). With SVM-PTFs, the smallest RMSE was recorded when was used as single input data (RMSE = 0.026 cm 3 cm-3). As for R 2 of SVM-PTFs, it was the highest with and SC as input data (R 2 = 0.84). The SLR-PTFs using as single predictor after stratification by texture performed better (RMSE = 0.031 cm 3 cm-3 and R 2 = 0.88) than the ANN-PTFs using one or two soil characteristics as input data. Comparison of SLR-PTFs with SVM-PTFs showed that the latter performed slightly better than SLR-PTFs after stratification by texture but R 2 was smaller when was used as the single predictor. Use of a predicted value of the bulk density at field capacity to obtain a value of in situ volumetric water content at field capacity led to poorer performances of the SLR-PTFs but after stratification by texture they remained close to those recorded with ANN-PTFs or SVM-PTFs when they used a single soil characteristic as input data. Finally, our results showed that associating OC to the input data did not increase the perfomances of the ANN-PTFs and SVM-PTFs.