Compared with the standard BP decoding, the BP decoding based on LLR reduces the computation burden, but is still complex. As a result, this may consume lots of hardware resources. To solve this problem, this article proposes an improved LLRBP decoding algorithm with dynamic bit node selection. According to the recursion of extrinsic messages and the satisfaction of parity equation, the part of nodes whose belief is the lowest is selected dynamically to participate in next iteration and update their extrinsic message. At the same time, the other variable nodes don't need to update their extrinsic messages in next iteration. The simulation result shows that the loss of the performance of advanced algorithm in signaltonoise is 0.1 dB and 0.07 dB compared with standard LLRBP decoding algorithm, when the codelength is 1 024 and 2 048 at 104 biterror rate and at 0.5 code rate.