This is the first comprehensive study on the optimization of seed distribution in a crystallization process. For a batch crystallizer, a dynamic programming formulation optimizes a property of the product crystals over the supersaturation profile and the seed characteristics, namely the mean size of the seed crystals, the seed mass, and the width of the seed distribution. Three optimization objectives are considered: (1) weight mean size, (2) coefficient of variation, and (3) the ratio of the nucleated crystal mass to seed crystal mass. Different objectives lead to substantially different optimal seed distributions. It is shown that optimizing over the seed, distribution can have a larger effect on the product crystal size distribution than optimizing over the supersaturation profile.