The cascade channel estimation of Amplify-and-Forward (AF) relay system was dealt with by using Bayesian framework. According to the maximum a posteriori (MAP) and the minimum mean of square error (MMSE) rules, two estimators were derived combining the received training signal with the probability distribution function (PDF) knowledge that cascade channel comprises two independent Rayleigh channels. Due to the intractability of exact Bayesian Cramer-Rao Bound (CRB), alternatively, the approximated Bayesian one was obtained based on the principle of maximum entropy to compare with the CRB of MLE. The numeral simulation demonstrates that the new MAP estimator can bring about significantly improved performance, especially in the low signal-to-noise ratio regime.