Abstract The paper presents a methodology for pre-processing the combustion time intervals, that is the basic signal used in misfire detection strategies, with the aim of increasing the signal-to-noise ratio to enable a more efficient misfire diagnosis, especially when the engine is running at high speeds and low loads. The performance of the basic misfire detection algorithm shows that in those engine operating conditions the background noise amplitude has approximately the same value of the information related to the misfire presence, thus hiding the misfire event that may not be detected. The proposed methodology is based on the correction of the combustion time signal cycle-by-cycle, using a vector of data that take into account the specific behavior of every cylinder. The vector of data for the combustion time correction is stored in a map inside the control unit and could be continuously updated with an auto-adaptive learning technique.