In the context of functional electrical stimulation, neural recording is one of the main issues. For instance, the control of the limbs in people with motor deficiencies needs information about the muscle lengths and speeds that can be extracted from electroneurograms (ENG) carried on afferent peripheral nerves. The aim of this study is to propose an non-invasive and spatial-selective electrode (because specific informations are carried into different fascicles). To do so, we investigate the spatial properties of an extracellular action potential (AP). This properties are described qualitatively and quantitatively using analytical study on an inhomogeneous an anisotropic nerve model. Then, a spectral analysis on this spatial signal discriminates the different frequency components. Low spatial frequencies represent the global shape of the signal, whereas high frequencies are related to the type of fibers. We show that the latter is rapidly attenuated with the distance and thus, being a local phenomenon, can be used as a selective measurement. Finally, we propose a spatial filtering based on electrode design and an electronic architecture to extract this high frequencies.