Water sources used for plant identification coupled with stable isotopes are essential to improving the understanding of eco-hydrological processes and ecological management in water-limited ecosystems. Many approaches associated with stable isotopes have been used to determine plant water source apportionment. However, inter-comparisons of different methods are still limited, especially for Bayesian mixing models. In this study, we tested linear mixing models (IsoSource) and Bayesian models (SIAR, MixSIR and MixSIAR) to identify sources of water absorbed by Vitex negundo and Sophora viciifolia (shrubs) and Artemisia gmelinii (subshrub) during the growing season in the semiarid Loess Plateau. The results showed that there was no significant difference in the predicted plant water source fractions using only stable hydrogen isotope (δ2H) and only stable oxygen isotope (δ18O) with the IsoSource model. No significant difference was found in plant water source apportionment by the three Bayesian mixing models combined with δ2H and δ18O except for individual months. The SIAR and MixSIAR models detected no pronounced seasonal variations in plant water uptake, while the MixSIR model did detect seasonal variations. Overall, the SIAR and MixSIAR models exhibited relatively better water source apportionment performances than that of the MixSIR model. This discrepancy may be attributed to the difference in the post distribution simulation algorithm. This study provides critical insights into choosing a suitable method for identifying plant water source apportionment in arid and semiarid regions.