Abstract This paper is concerned with the Indian design of a 220 MWe pressurized heavy water reactor (PHWR) having natural uranium (NU) fuel and heavy water as moderator and coolant. At the beginning of life, it is necessary to flatten the power by loading some depleted uranium (DU) bundles to achieve a nearly full power operation. The determination of best possible locations of DU bundles, which maximize fuel economy as well as safety, is a large-sized combinatorial optimization problem with constraints. In the past, 384 DU bundles have been loaded in locations determined by manual intuition in an Indian PHWR and maximum permissible power of 93% full power (FP) was obtained. In the present paper, a modern evolutionary algorithm called estimation of distribution algorithm (EDA) is used to improve upon this distribution. Optimum distributions of DU bundles which maximize K eff and give 100% FP without violating safety parameters such as maximum permissible bundle power, channel power, channel outlet temperature and permitted reactivity worths of shut-down systems are obtained. Another aspect studied in this paper is to find out how far one can increase the number of DU bundles loaded in the core. This will minimize the NU bundles requirement, extract more power from DU bundles and thus provide better fuel utilization. The idea is to conserve NU bundles. The optimum distribution of DU bundles has been obtained for the total number of DU bundles ranging from a few hundreds to a few thousands. It is found that, depending on various conditions, about 60–80% of the core can be loaded with DU bundles leading to a substantial saving in NU bundles. Some variation in the implementation of EDA to generate loading pattern of PHWR reactor core is also studied.