The performance of particle-in-cell (PIC) programs executing on Cray X-MP supercomputers is highly dependent on a critical region of the code: the collection of grid quantities from particle quantities, which cannot directly be vectorized. To this purpose methods allowing for either partial or total vectorization have been proposed; these methods either require very large additional storage or are limited to processing short vectors. After the illustration in three dimensions of the different methods devised and used up to now, a new scheme for the vectorization of the interpolation on the grid is proposed. This method requires a small amount of extra storage, is completely vectorized, and makes it possible to take full advantage of the intrinsic vectorizability of the physical system. A comparison using a tridimensional example, representing a vast range of simulations, shows a relevant speedup for the scheme proposed over the other methods in existence up to now, up to a factor of 3 when compared to the original scalar code.