In the context of functional electrical stimulation it is of high interest to have an objective measure of the effect of stimulation and to get a sensitive feedback from afferent neural signal. Many studies aim to detect or record this information from inside the nerve but they have to overcome several difficulties. In chronic implants, measure of the nerve activity has to be little invasive making the use of a cuff electrode very suitable. The main issues with this kind of electrode are parasitic noise and poor selectivity of the recorded signal. In this paper, we propose an optimized layout for multipolar recording electrode. The main idea is to find the best value for the inter-pole distance and the most relevant processing in order to both improve selectivity in the nerve and to reject external parasitic signals. In this study, we put emphasis on simulation of action potential as a method to help the electrode specification. The amplitude of the expected signal is evaluated in both spatial and frequency domains, with respect to axons variability. Then, the selectivity of the proposed design is compared to state-of-art electrode layout. The proposed design, with the associated pre-processing shows a real improvement of the electrode selectivity and a very efficient parasitic signal rejection. The drawback is a decrease of the sensitivity that nevertheless remains compatible with integrated micro-circuit amplifiers.