Post selection inference is of particular interest in medical trials and other research studies which require the selection of treatments. The empirical Bayes estimator of the selected treatment mean dominates the usual estimator with respect to squared error loss. However, the naive empirical Bayes confidence interval (EBCI) fails to achieve its target coverage probability. This talk will propose a construction of EBCI for the selected mean which is second order bias adjusted and achieves the target coverage asymptotically.