To solve the low estimation accuracy of MQAM signal-to-noise (SNR) estimate under the low SNR condition, an improved SNR estimation algorithm based on comprehensive utilization of higher-order statistics is proposed. According to the difference between the topmost rank of the higher-order statistics, the linearity relationships of three kinds of SNR and various higher-order statistics expression are created, which are converted to three kinds of whole regression patterns with a whole regression linear analysis method, and the pattern coefficients are solved. It can fully use the useful information of the higher-order statistics, and increase the accuracy of SNR estimate. Simulation results show that under the low SNR condition the new algorithm can get less SNR estimating errors and obviously better estimation performance than traditional algorithms. Moreover, the estimation performances of three kinds of patterns increase in turn, and it can carry on a selection to three kinds of patterns according to the different SNR request.