Cuff electrodes have several advantages for in situ recording ENG signal. They are easy to implant and not very invasive for the patient. Nevertheless, they are subject to background parasitic noise, especially the EMG generated by the muscles. We show that the use of cuff electrodes with large numbers of poles can increase their sensitivity and their selectivity with respect to an efficient noise rejection. We investigate several configurations and compare the performances of a tripolar cuff electrode versus a multipolar one in numerical simulation. One the other hand the use of cuff electrodes leads to the recording of the sum of the signals generated by all the axons within the nerve. This puts in evidence the need of signal separation techniques that require a great quantity of information. Again, we show that multipolar electrodes can solve this problem since poles can be switched one to another, provided that they are distributed along a regular tessellation. Finally, we present the structure of an ASIC preamplifier processing a spatial filtering to obtain the Laplacian of the potential rejecting low-frequency noise.