This paper deals with soft decision based decoding schemes for channel quality information (CQI) in long-term evolution (LTE) systems. We first propose a new soft decision decoding method based on modified Reed Muller decoding (MRMDSD) to reduce computational complexity of maximum-likelihood decoding (MLD). In addition, we provide the theoretical upper bound of channel code of CQI on additive white Gaussian noise (AWGN) channels. The upper bound yields an accurate expression for the error probability. Simulation results show that the proposed MRMD-SD approximates to the MLD performance while reducing computational complexity in multiplications by 97.5%.