The TanDEM-X DEM is a valuable data source for estimating glacier mass balance. However, the accuracy of TanDEM-X elevation over glaciers can be affected by microwave penetration and phase decorrelation. To investigate the bias of TanDEM-X DEMs of glaciers on the Tibetan Plateau, these DEMs were subtracted from SPOT-6 DEMs obtained around the same time at two study sites. The average bias over the studied glacier areas in West Kunlun (175.0 km2) was 2.106 ± 0.012 m in April 2014, and it was 1.523 ± 0.011 m in Geladandong (228.8 km2) in October 2013. By combining backscatter coefficients and interferometric coherence maps, we found surface decorrelation and baseline decorrelation can cause obvious bias in addition to microwave penetration. If the optical/laser data and winter TanDEM-X data were used as new and historic elevation sources for mass-balance measurements over an arbitrary observation period of 10 years, the glacier mass loss rates in West Kunlun and Geladandong would be potentially underestimated by 0.218 ± 0.016 and 0.158 ± 0.011 m w.e. a−1, respectively. The impact is therefore significant, and users should carefully treat the bias of TanDEM-X DEMs when retrieving a geodetic glacier mass balance.