The China Seismo-Electromagnetic Satellite (CSES) was successfully launched in February 2018. The high precision magnetometer (HPM) on board the CSES has captured high-quality magnetic data that have been used to derive a global lithospheric magnetic field model. While preparing the datasets for this lithospheric magnetic field model, researchers found that they still contained prominent residual trends within the magnetic anomaly even once signals from other sources had been eliminated. However, no processing was undertaken to deal with the residual trends during modeling to avoid subjective processing and represent the realistic nature of the data. In this work, we analyze the influence of these residual trends on the lithospheric magnetic field modeling. Polynomials of orders 0–3 were used to fit the trend of each track and remove it for detrending. We then derived four models through detrending-based processing, and compared their power spectra and grid maps with those of the CSES original model and CHAOS-7 model. The misfit between the model and the dataset decreased after detrending the data, and the convergence of the inverted spherical harmonic coefficients improved. However, detrending reduced the signal strength and the power spectrum, while detrending based on high-order polynomials introduced prominent distortions in details of the magnetic anomaly. Based on this analysis, we recommend along-track detrending by using a zero-order polynomial (removing a constant value) on the CSES magnetic anomaly data to drag its mean value to zero. This would lead to only a slight reduction in the signal strength while significantly improving the stability of the inverted coefficients and details of the anomaly.