We investigated the temporal structure of parallel spike trains. The spike trains were reconstructed from the simultaneous activity of 4 to 28 extracellularly recorded hippocampal pyramidal cells during different behavioral conditions. We found evidence for the generation and persistence of stable spatio-temporal spike patterns (sequences) within 200 msec time window. The repetitive temporal structure of the spikes were extracted by different sequence detection methods. The statistical and behavioral significance of spike sequences was also investigated. Three findings will be discussed ( 1) Using Monte Carlo statistics and joint probability mapping method we observed sign ificantly more repetitions than expected based on different spike train statistics, consistently, across all animals investigated. Spike sequences were stable during the same behavioral state. The order of 3 to 5 spikes at different cells as well as the delay between spikes were invariant within up to 200 msec with a precision of 6 to 20 msec. The possibility that sequences are artifacts of the periodic modulation was ruled out. (2) Long (50-200 msec) sequences were associated with theta activity. In contrast, short (<50 msec) sequences were coincident with sharp wave-related ripple activity. The same sequences which were generated during theta activity recurred with a 4 to 8 fold temporal compression during subsequent sharp wave episodes. (3) Spike sequences generated during active running behavior were more likely to recur during the subsequent sleep phase than other sequences. Sequences of the active behavior of the rat were reflected by the subsequent sleep state more than by the preceding sleep phase. Sequences are, therefore, hypothesized to represent association of cells within a functional assembly that may support memory function in the hippocampus.