Abstract: The water stress threshold is one of the most important indexes to evaluate the response of crops to water stress. Water control in the pot culture is also a commonly-used way to simulate drought stress conditions. The absorption of soil water by crops is mainly decided by the soil water potential. And the relationship between soil water content and water potential varies with matric medias. However, the water stress threshold of a crop is usually given in the form of mass water content. However, there is some inconsistency in the matric media under previous pot experiments. This study aims to rapidly obtain the water retention curves of base materials in the pot experiments, in order to predict the potting soil water retention curves. The peat soil, vermiculite, and perlite were taken as matric materials in the pot experiments. The water retention curves of various media were measured, including three kinds of matric materials and eight different proportional compositions. The van Genuchten model was used to fit these water retention curves. After that, the pedotransfer functions (PTFs) of water retention curves were established using multiple regression and artificial neural network. The normalized root mean square error (SNRMSE), determination coefficient (R2), and mean error (SMR) were selected as the accuracy indicators of the model. Furthermore, the model stability was verified through 30 cycles to finally determine the optimal modeling and the predicted water retention curve. The results showed that an ideal fitting effect of the van Genuchten model was achieved in the water retention curves of (R2>0.99). There were some significant differences in the water retention curves of the three potting medias. Peat soil presented more invalid pores than vermiculite and perlite, resulting in a large water content in the high-water suction section (water suction>100 kPa). The high capillary pore content of vermiculite led to a high-water content in the middle and high-water suction sections (water suction 33-800 kPa). Pearlite shared the higher permeability and low water holding capacity. Water was rapidly released with the increase of water suction. The prediction accuracies of the artificial neural network model for the peat soil water retention curves and vermiculite media materials were better than the multiple regression model. However, the opposite trend was observed for the peat soil and perlite media materials. Two different prediction models had different errors for the water retention curve parameters in the different media materials combinations. The multiple regression models often underestimated the prediction of saturated water content. As such, the multiple regression models underestimated the water content of media materials, compared with artificial neural networks under different water suction. The artificial neural network with PTFs reached a higher accuracy than the multiple regression with PTFs. The accuracy of multiple regression with PTFs had smaller variation and higher stability (smaller SNRMSE, larger R2, and mean error closer to 0), compared with the artificial neural network. Meanwhile, there was a large variance and distribution range of SNRMSE, R2, and SMR of PTFs developed by artificial neural networks, compared with the multiple regression. Therefore, better stability was achieved in the PTFs developed by multiple regression. In terms of accuracy and stability, multiple regression can be selected as the best to develop the water retention curve PTFs in various plot experiment medias. The finding can provide a strong reference for the rapid acquisition of water retention curves using the comparison between the pot water control experiments.