Series of water-soluble anions in soils can be used to identify the types and degrees of soil salinization. This study aimed to determine the optimal model for estimating soil soluble anion content by multispectral remote sensing. Based on Sentinel-2B and Landsat8-OLI multispectral data and the contents of anions in saline soil in Pingluo County, northern Ningxia of northwest China, the sensitive spectral data were screened by Pearson correlation coefficient (PCC) and gray relational analysis (GRA). A soil anion content estimation model based on support vector machine (SVM) was constructed. The feasibility of remote sensing images to estimate soil anions in arid and semi-arid regions was determined by comparison with partial least squares regression (PLSR). The results showed that: (1) Cl^- was the highest soil anion, followed by SO₄^2-. The Cl^- content showed strong variation. (2) With PCC and GRA as variable screening methods, no sensitive variables of SO₄^2- and HCO₃^- were selected. Cl^- satisfied the conditions and there were multiple sensitive spectral indicators. (3) Overall estimation efficacy of Cl^- model constructed by PCC based variable screening approach was better than GRA. (4) For the two inversion models of Cl^- content, the SVM model showed a higher accuracy than PLSR model. For the single-phase images, the best inversion performance for Cl^- estimation was achieved with the PCC-SVM model constructed by the sensitive band and salinity index selected by Sentinel-2B, with the values of coefficient of determination (R_P²) and relative percent deviation (RPD) being 0.989 and 6.616. For the Sentinel-2B and Landsat8-OLI combined images, PCC-SVM model constructed with the sensitive bands achieved the best inversion effect, with the values of R_P² and RPD being 0.895 and 2.066. Our results indicated the feasibility of quantitative estimation of Cl^- based on Sentinel-2B satellite data, providing a scientific basis for the rapid identification of soil salinization in local and similar areas.