Field capacity (FC) and permanent wilting point (PWP) are important physical properties for evaluating the available soil water storage, as well as being used as input variables for related agro-hydrological models. Direct measurements of FC and PWP are time consuming and expensive, and thus, it is necessary to develop related pedotransfer functions (PTFs). In this study, stepwise multiple linear regression (SMLR) and artificial neural network (ANN) methods were used to develop FC and PWP PTFs for the deep layer of the Loess Plateau based on the bulk density (BD),sand, silt, clay, and soil organic carbon (SOC) contents. Soil core drilling was used to obtain undisturbed soil cores from three typical sites on the Loess Plateau, which ranged from the top of the soil profile to the bedrock (0–200 m). The FC and PWP were measured using the centrifugation method at suctions of − 33 and − 1500 kPa, respectively. The results showed that FC and PWP exhibited moderate variation where the coefficients of variation were 11 and 23%, respectively. FC had significant correlations with sand, silt, clay, and SOC (P < 0.01), while there were also significant correlations between all of the variables and PWP. In addition, sand was an important input variable for predicting FC, and clay and BD for predicting PWP. The performance of the SMLR and ANN approaches was similar. In this study, we developed new PTFs for FC and PWP as the first set of PTFs based on data obtained from deep profiles in the Loess Plateau. These PTFs are important for evaluating the soil water conditions in the deep profile in this region.