Recording and stimulating from the peripheral nervous system are becoming important components in new generations of bioelectronic systems. Neurostimulation has seen a long history of success in chronic application but robust neural recordings remain a challenge. Nerve cuff electrodes are an option that have demonstrated chronic viability in humans but lack the required recording selectivity. The objective of this thesis was to increase the recording selectivity achievable using multi-contact nerve cuff electrodes. Neural activity recorded from multi-contact nerve cuffs for a given neural pathway will have a distinct spatiotemporal signature. This signature is determined by the location of the bioelectric sources in the nerve combined with the type of fiber, which governs the propagation properties of the action potentials. Both the spatial and temporal information are needed to interpret the physiological meaning of the recorded activity, but have not been adequately combined in signal processing techniques to date. We investigated selective recording approaches based on spatiotemporal characterizations of neural pathways. First, we performed a simulation study to investigate the effects of the spatial and temporal information on recording selectivity. A matched filter approach was developed and results showed that neural pathways that had different fiber types were highly distinguishable, whereas neural pathways with similar fiber types showed varying distinguishability results depending on pathway locations. Next, we performed a validation study in vivo in a rat sciatic nerve model. We tested our matched filter approach and machine learning algorithms in distinguishing the neural pathways. The results of the matched filter approach were comparable to the simulation study, validating our simulations, and the machine learning algorithms showed improvements in recording selectivity. These results demonstrated classification of individual naturally evoked compound action potentials (nCAPs) for the first time. Lastly, we demonstrated that the use of a convolutional neural network (CNN) led to further improvements in recording selectivity. With the help of a recurrent neural network, joint angles could be tracked accurately based on the classified nCAPs. These findings suggest that nerve cuff electrodes can provide robust recordings and be beneficial in many neuroprosthetic and neuromodulation application.