Accurate snow cover information is of great significance to the study of meteorology, hydrology, and global climate change. Remote sensing techniques play an important role in large-scale and high-frequency snow cover monitoring. Nowadays, SNOMAP algorithm is the most common method for remote sensing monitoring of snow, which mainly uses fixed NDSI (Normalized Difference Snow Index) thresholds to identify snow. However, this method ignores the temporal variations of snow spectral information, leading to monitoring errors of snow cover. In this study, we proposed an adjusted method to monitor snow cover by dynamic NDSI thresholds. This method adjusts fixed NDSI thresholds by using the average NDSI value of pure permanent snow as reference to reduce the influence of spectral fluctuations. Snow cover in the Sanjiangyuan area was identified and monitored by this method. There were four steps: (1) OLI and MODIS data of the same region, the same period and cloud-free were selected. The OLI NDSI threshold of the best snow cover recognition was determined by human-computer interaction. (2) The snow area monitored based on OLI data was used as the true value of the ground to calibrate the optimal MODIS NDSI threshold on the same day. (3) The average NDSI value of the pure permanent snow in the Sanjiangyuan area on the same day was counted. The elevation of the pure permanent snow pixels was more than 5800 meters and the FSC (Fractional Snow Cover) of them was 100%. (4) The functional relationship between the optimal MODIS NDSI threshold and the average NDSI value of the pure permanent snow was established. The dynamic MODIS NDSI threshold was obtained by the linear regression and varied with the average NDSI value of pure permanent snow. Results show that: (1) Based on daily MODIS data, there was a good linear relationship between the optimal NDSI threshold for snow cover monitoring and the average NDSI value of pure permanent snow on the same day, and the determinant coefficient R² reached 0.86. (2) The dynamic NDSI thresholds of Sanjiangyuan area were between 0.29 and 0.37, and the average value of NDSI threshold was about 0.33, indicating that 0.40 as the NDSI threshold would underestimate the snow cover area of the Sanjiangyuan area. (3) The average values of the approximation ratio, the overall classification accuracy, and F of dynamic NDSI threshold method were 96.61%, 94.62%, and 91.99%, respectively. Compared with the monitoring method with the fixed NDSI threshold of 0.33, they were improved by 5.17%, 0.70%, and 1.14%, respectively. Our findings demonstrate the effectiveness of the proposed method.