Multi-level cell (MLC) threshold voltage is seriously affected by random telegraph noise (RTN), data retention noise (DRN) and cell-to-cell interference (CCI) in MLC flash channel, which leads to lower-accuracy of the log-likelihood ratio (LLR) value and affects the performance of low-density parity check (LDPC) codes under soft decision decoding in MLC flash memory. Based on the analysis of the error characte ristics of MLC flash memory, by using the entropy function of MLC threshold voltage for calculating the overlap region of the adjacent threshold voltage distribution to determine the reliability value of stored bit corresponding to MLC threshold voltage, a strategy of dynamically updating the LLR value for stored bit is designed. Thus, an improved min-sum decoding algorithm of LDPC codes for MLC flash memory is presented over the RTN, DRN and CCI noises. The simulation results show that the decoding performance of the improved min-sum decoding algorithm for MLC flash memory is better than that of the conventional min-sum decoding algorithm. Furthermore, the average number of iterations is reduced.