Many studies of neuronal activity require isolation of the extracellular wave form (spike) of a single neuron from the potentials generated by nearby cells. A variety of methods for spike sorting exists, but most are expensive and require specialized hardware and software. Moreover, there is no easy and objective way for evaluating and comparing the performance of spike sorting devices. We describe here a system for on-line spike sorting that is implemented on an IBM PC/AT computer using commercially available hardware and C-language software. Spikes are detected after crossing an amplitude threshold and are sorted or rejected by template matching. The templates are constructed in a learning phase, using a fast manual sorting of all detected spikes. Later, each detected spike is matched against all defined templates. A detected spike which does not match any template, or matches more than one, is rejected. A continuous display of the wave forms of the last 256 sorted, double-matched, and rejected spikes is used as the main tool for parameter adjustment and error detection. Also described is a new and highly versatile tool for generating appropriate wave forms for critical evaluation of sorter performance. Using the same hardware and software tools, a simulation program mimics the extra-cellular activity of several neurons by linear combination of two vectors and added random noise. The size, shape and the variability of the action potential, as well as its firing pattern, can be adjusted. Comparison of the sorter output with the known simulated activity is used to examine the sorter performance and limitations.