Abstract: SNR maximum likelihood estimator based on data directed decision was evaluated first, and was concluded as a biased estimator. An unbiased estimator can be acquired through high order moments of received data. Two SNR estimators based on iteration method were developed in this paper. One is solving the QPSK NDA maximum likelihood equation by gradient approximation, which has better convergence performance than other methods. The other is based on the statistical parameters of received data, which not only makes received data buffering unnecessary, but also reduces iteration calculation greatly compared with other iteration methods, and more important, not at the cost of estimation performance degradation. Finally, the result of several SNR estimators in performance and calculation were compared and analyzed.