When extracellular action potentials (spikes) from cultured neurons are recorded using microelectrode arrays in open wells, their amplitudes are usually quite small (often below the noise level) despite the extracellular currents originating from the relatively large surface area of neural cell somata. In this paper rat cortical neurons were seeded into one well of a two well system separated by 3×10 μm microtunnels and then 7 days later into the second well forming a feed-forward network between two small neuronal assemblies. In contrast to measurements in the open well spikes recorded from axons within the restricted volumes imposed by the microtunnels are often several orders of magnitude larger than in the open well, with high signal to noise ratio, despite the currents originating in the much smaller surface area of the axon. Average signal amplitudes exceeding 250 μV are typical, with some signals as large as 4.5 mV (signal-to-noise ratio up to 450), twenty times greater than the maximum recorded from electrodes in adjacent but open wells. We confirm the dependence of signal amplitude on the impedance of the microtunnel and discuss possible reasons for the phenomenon.