We report a new approach to the retrieval of ice particle temperature in the atmosphere from midinfrared remote sensing observations. Two different in physical nature spectral bands of water ice, the libration band near 850 cm−1 and O-H stretch band at 3200 cm−1, are used simultaneously. The use of both bands in the unified least squares fitting procedure enables one to enhance the stability of fitting, as well as to improve the reliability and accuracy of retrieved ice temperature values. A set of reference spectra for the least squares minimization is compiled from experimental temperature-dependent optical constants for water ice that are available in the literature. The method is applied to the observations made by the Atmospheric Chemistry Experiment Fourier Transform Spectrometer (ACE-FTS) instrument on board the SCISAT-1 satellite. According to our results, a combination of the vibration (3200 cm−1) and rotation (850 cm−1) bands within a unified minimization procedure can significantly, by about 4K, reduce the upper estimate of the total uncertainty in the retrieved temperatures from 12 K down to 8 K. The average temperature of bright mesospheric ice clouds observed by ACE-FTS during July 2005 is found to be about 131K, in contrast to the value of 135K retrieved from the vibration band alone.