Multi-unit activity recorded from two electrodes positioned at a distance on a nerve may be analysed by cross-correlation, but units similar in direction and velocity of propagation cannot be distinguished and separately evaluated by this method. To overcome this limit, we added two features, represented by the impulse amplitudes of the paired recordings, to the dimension given by the impulse delay. The analysis was fractionated according to the new dimensions. In experimental recordings from the locomotor appendage of the lobster Homarus americanus, the fractionated analysis proved capable of identifying the contributions of single active units, even if these were superimposed and indiscernible in the global cross-correlation histogram. Up to 5 motor and 10 sensory units could be identified. The shape of the paired impulses was evaluated by an averaging procedure. Analogous evaluations on simulated recordings made it possible to estimate the influences exerted on performance by variations in noise level and in the number and firing rate of active units. The global signal could be resolved into single units even under the worst conditions. Accuracy in evaluating the amount of unit activity varied, exceeding 90% in about half of the cases tested; a similar performance was attained by evaluation of the impulse shapes.