The empirical Bayes prediction of population total in a finite population with normal inverse-Gamma priors is investigated. Firstly, the Bayes prediction for population total is obtained by a new method. On the basis, the empirical Bayes predictions are proposed. Secondly, the performances of empirical Bayes prediction are analyzed by two cases. The result showed that empirical Bayes prediction is asymptotically optimal and can be proved to be the best linear unbiased prediction under the prediction mean squared error. Finally, a simulation study is used to illustrate the proposed results.