In this paper, a new replacement policy for a deteriorating item from a mixed population is suggested. The item is randomly selected from a mixed population composed of two stochastically ordered subpopulations and the corresponding subpopulation from which the item is selected is unknown. As the operational history contains the information about the subpopulation from which the item is selected, it is reasonable to employ this information in determining the replacement policy. In the new replacement policy, the age at which the item will be replaced is not determined before the operation of the item, but it is determined based on the failure/repair history of the item observed during the initial operation. It is shown that the proposed replacement policy outperforms the ordinary periodic replacement policy. The properties of the optimal replacement policy will be studied in detail.