Abstract: The Xin'anjiang model stands as one of the most important hydrological models. This paper develops a differentiable Xin'anjiang model coupled with the CemaNeige module（dMXAJ）under the framework of differentiable parameter learning. Specifically, the long short-term memory（LSTM）network is employed to generate parameters for the dMXAJ through forward propagation, using catchment attributes and meteorological data as inputs. These parameters are then fed into the dMXAJ and optimized via backpropagation. Four models are used to investigate the effectiveness of dMXAJ across 531 catchments from the Catchment Attributes and Meteorology for Large-sample Studies. The results show that the CemaNeige module effectively improves the performance of the Xin'anjiang model, increasing the median Kling-Gupta efficiency（E_KG）from 0.58 to 0.68. For the local dMXAJ, the median E_KG is improved to 0.70; for the regional dMXAJ, the median E_KG is improved to 0.72. These improvements can be attributed to the consideration of snow accumulation and melting, the effectiveness of the differentiable parameter learning and synergistic effects. Overall, the differentiable parameter learning effectively facilitates the development and application of the Xin'anjiang model coupled with the CemaNeige module.