Salinity is an important parameter to characterize the physical properties of water bodies. In order to explore the ability of remote sensing data to measure the salinity of low-salt lakes, this study used remote sensing reflectance data in blue, green, red, and near-infrared bands with 10 m spatial resolution of sentinel-2 to analyze the relationship between measured surface salinity and reflectance in the Co Ngoin Lake. Based on different combinations of variables, which included the reflectance of four bands and normalized difference water index (NDWI), a linear regression model was constructed with salinity data, and the accuracy evaluated. The model was also used to invert the surface salinity of the Co Ngoin Lake. The results showed that the correlation between reflectance and salinity in the green band was higher than that in other bands. However, the near-infrared band reflectivity had the highest correlation with salinity, when the salinity was lower than 3. Among the 9 models with the combination of variables, the addition of the NDWI variable improved the accuracy of model inversion, which was higher than the accuracy of the model without the NDWI variable. The linear regression model of the three variable combinations of NDWI, near-infrared band and blue band had the highest accuracy of salinity inversion. The mean absolute error (MAE) was 0.103, and the salinity observed value and predicted value correlated well. In the combination without NDWI variable, the salinity inversion accuracy of the green band and red band was high, with MAE of 0.126. The spatial distribution of water surface salinity of the Co Ngoin Lake generally presented a spatial pattern of low salinity at the shore and estuary, and high and relatively uniform distribution within the lake. From the inversion results, the average salinity of the Co Ngoin Lake was approximately 4.14, which was very close to the measured average (4.15). The results verified the effectiveness of the multispectral remote sensing method, which has the advantages of being fast, convenient and highly accurate in predicting the water surface salinity of the Co Ngoin Lake. This study has guiding significance for low lake surface salinity inversion using high-resolution multi-spectral remote sensing data. In addition, it has significant value for the protection and sustainable use of aquatic biological resources.