The research described in this paper was aimed at determining any common precedents for abnormal combustion events in a gasoline engine in order that their occurrence might be predicted and ultimately corrected. Combustion data was collected for 1200 cycles from a motorbike spark ignition engine running at 4,000rpm and 1.15 bar BMEP at fuelling conditions ranging from rich through stoichiometric to lean. The least cyclic variability occurred during slightly rich fuelling and was best characterised through the heat released per cycle. Phase-lag plots of the heat released per cycle showed boomerang shaped patterns, indicating deterministic cyclic variability. This suggested that an algorithm capable of predicting poor cycles could be developed. It was found, however, that the majority of poor combustion events are preceded by cycles of an approximately average heat release and that poor combustion events are likely to be followed by a strong combustion event. Importantly not all poor cycles follow this pattern. This means that existing prediction algorithms which are based on preceding cycles would not be suitable in this case. An examination of the chain length of good or average combustion events between poor events did not show any chain length commonality across the AFR’s. The findings of this paper support the view that the deterministic patterns in the cyclic variations in combustion are present but that they are complex and cannot be predicted using simple metrics based on neighbouring cycles. A more thorough understanding of the fundamental contribution of previous combustion cycles to the variability is required for predictive tools to be effective.