A 56-contact matrix nerve cuff electrode (seven rings with eight contacts each) was used to obtain recordings from the rat sciatic nerve, which were then discriminated as originating from one of three fascicles (tibial, peroneal, and sural branches). The influence of the number and location of the recording contacts on the classification accuracy was studied. The performance of a classifier was shown to be superior when data was available from all 56 contacts, compared to when only the eight contacts of the middle ring were used (as in previously proposed multicontact tripolar cuff designs). By examining the performance variations as contacts were included one at a time (in order of decreasing positive impact on performance), it was further shown that the matrix configuration could outperform the single-ring configuration with only a small number of contacts. We can therefore conclude that the performance improvement is not due to the sheer number of contacts, but rather to the possibility of selecting the most informative locations around the nerve. The results could have important implications for the design and use of multicontact nerve cuff electrodes.