Burn-in is a method used to eliminate the initial failures in field use. In this paper, we will consider an information-based burn-in procedure for repairable items, which is completely new type of burn-in procedure. By this procedure, based on the operational (failure and repair) history of the items observed during burn-in procedure, those with poor reliability performance are eliminated. From a probabilistic point of view, this burn-in procedure utilizes the information contained in the 'random paths' of the corresponding point processes. A general formulation of the model will be suggested, and under the suggested framework, two-stage optimization procedure for determining optimal burn-in procedures will be studied in detail.