The detection and classification of extracellular action potentials (i.e. spike) of various single neurons from extracellular recordings are crucial for extracting neuronal spike sequences and thereby for investigating the mechanisms of neural information processing in the central nervous system. In order to increase the correctness of spike detecting and sorting, a new analysis algorithm for processing multi-channel spike signals recorded from rat hippocampi with silicon microelectrode arrays is presented. Four recording contacts on the electrode array are arranged close enough to simultaneously record spikes emitted from same neurons. Firstly, the algorithm extracts all spikes in the four channel recordings by using a multi-channel threshold detection method. Secondly, the algorithm classifies the spikes based on a principle component analysis for a specifically designed type of compound spike waveforms. The compound spike waveform is formed by linking four spike waveforms of a same neuronal firing in the four recording channels one by one in series. The test results with both synthetic datasets and experimental recordings reveal that compared with corresponding traditional single-channel algorithm, the multi-channel algorithm can significantly enhance both the number of extracted spikes and the correctness of spike classifications. The algorithm can also increase the number of isolated neurons from a single experimental preparation. These results indicate that the novel method is efficient for the automatic detection and classification of neuronal spikes.