Brain machine interfaces with chronically implanted microelectrode arrays for signal acquisition require algorithms for successful detection and classification of neural spikes. During the design of such algorithms, signals with a priori known characteristics need to be present. A common way to establish such signals is to model the recording environment, simulate the recordings and store ground truth about spiking activity for later comparison. In this paper, we present a statistical method to expand the spike libraries that are used in a previously presented simulation tool for the purpose described above. The method has been implemented and shown to successfully provide quick access to a large assembly of synthetic extracellular spikes with realistic characteristics. Simulations of extracellular recordings using synthesized spikes have shown to possess characteristics similar to those of in-vivo recordings in the cat cerebellum.