This article sets out to indicate the type of description that can be obtained for early combustion in a spark ignition petrol engine using a combination of multi-variate data obtainable from modern multi-processor controlled engine test systems and recent developments in image analysis techniques. The image processing is applied to sequences of pictures of the chemi-luminescent combustion light captured electronically at 0.1ms intervals during each cycle. This approach is well suited to studies of the genesis of cyclic variability with the conventional engine variables being used to identify sub-classes of combustion. By combining straightforward object size analysis with the new technique of Dilation Correlation the paper shows how it is possible to separate out various elements of the early growth of the flame kernel. This separation allows the comparison of the purely thermal expansion aspect of flame kernel growth with the total flame growth. The techniques employed also yield the translational part of the flow-induced motion of the early flame kernel for the cycles under investigation. The emphasis of the work is focused strongly upon the approach and a very limited number of results are presented for purely illustrative purposes with no attempt being made at a comprehensive survey. The actual data were obtained from an unsophisticated side valve engine and are in any case unlikely to bear strongly on current trends in engine design. Given this poviso the results presented do indicate that under nominally uniform conditions there is a link between increasing uniformity of the various aspects of kernel growth and the cycle quality.