Pictorial databases require efficient and direct spatial search on the analog form of spatial objects and relationship instead of search based on some cumbersome alphanumeric encodings of the pictures. r-trees (two- dimensional b-trees) are excellent devices for indexing spatial objects and relationship found on pictures. their most important feature is that they provide high level object oriented search father than search based on the low level elements of spatial objects. this paper presents an efficient initial packing technique for creating r-trees to index spatial objects since pictorial databases are not update intensive but rather static, the benefits of this technique are very significant.