Abstract: The security of LED, a lightweight block cipher proposed in CHES 2011, was evaluated by the template attack (TA). Several improvements of TA from the perspective of the preprocessing optimization was proposed. Firstly, the noise offset was calculated by using the phase-only correlation factor in the frequency view of the power trace to eliminate the data interference in the template building phase. Secondly, a novel character extracting method was proposed based on calculating the cross-cluster offset of different clusters classified by the plaintexts to cut the different leakage points from the power traces automatically. Thirdly, a dynamic effective power points choosing strategy was proposed by utilizing the mean value and the noises of the of power traces to evaluate the differences between different templates and improve the utilization of side channel information. Experiment results demonstrate that the proposed techniques of data alignment and automatically data cutting enlarge the differences of templates and reduce the number of the required power trace in both the template building and attacking phase. The proposed effective power points choosing strategy reduces the data complexity of the attack and only two power traces are required to launch the attack with the success rate of 100%.