We compare the performance of algorithms for automatic spike detection in neural recording applications. Each algorithm sets a threshold based on an estimate of the background noise level. The adaptive spike detection algorithm is suitable for implementation in analog VLSI; results from a proof-of-concept chip using neural data are presented. We also present simulation results of algorithm performance on neural data and compare it to other methods of threshold level adjustment based on the root-mean-square (rms) voltage measured over a finite window. We show that the adaptive spike detection algorithm measures the background noise level accurately despite the presence of large-amplitude action potentials and multi-unit hash. Simulation results enable us to optimize the algorithm parameters, leading to an improved spike detector circuit that is currently being developed.