Abstract: The LCS algorithm used in protocol feature recognition is a string matching algorithm to extract the longest string of continuous public substring. However, through theoretical analysis and some experimental situation, it can be seen that this process is a time complexity of higher computing process. If the input data packet is relatively large, the running time will be very long. To this end, the size and number of input packets have to be controled, which severely limits the size of the sample set. A GPU based method for accelerating the LCS algorithm was proposed. The CUDA platform was built and dedoyed and the parallel of LCS algorithm was researched on this platform. By the parallel study of LCS algorithm on the CUDA, the operation speed of the LCS is effectively enhanced. Highly competitive experimental results show that the LCS algorithm in the GPU is more effective and efficient than that in the CPU.