This letter examines the performance of two Advanced Microwave Scanning Radiometer-EOS (AMSR-E) ice temperature algorithms over first-year sea ice during the spring transition period where ice concentrations are close to 100%. The results showed, before snow melt, that the old AMSR-E algorithm overestimated the ice temperature by up to 18 K, which is relative to in situ and thermodynamically calculated snow/ice interface temperatures (T si's). An adjustment of vertically polarized ice emissivity of 6.9 GHz [εI(6V)] to 0.98, which was identical to the constant value used in the latest version of the AMSR-E ice temperature algorithm (posted July 2007), demonstrated a significant improvement in ice temperature retrieval. However, after snow melt, the ice temperature retrieval with any constant εI(6V) failed to correctly estimate the ice temperatures due to large variability in the physical properties of snow and, in turn, penetration depth and εI(6V). The results suggest that a local adjustment of εI(6V), which is by incorporating a simple thermodynamic model into the AMSR-E ice temperature algorithm, would be useful in improving the performance of the algorithm.