Abstract: An improved method and system are provided for the detection of internal combustion engine misfires utilizing optimum signal processing. The method and system exploit a measurement of engine crankshaft angular velocity in conjunction with hybrid electronic signal processing. Once the angular velocity signal is conditioned to minimize the effects of random error and external disturbances, four alternate computationally efficient methods may be used to extract information pertaining to individual cylinder torque productions. Two of the methods employ extremal samples of the estimated torque or velocity waveform to obtain a random torque nonuniformity index or metric. The other two methods utilize a transformation into the frequency domain after the input data is first sampled and windowed to substantially lower probability of error. All of the methods provide an M-dimensional torque non-uniformity vector of individual cylinder performance for each individual engine cycle.