Lean combustion in spark-ignition engines has long been recognised as a means of reducing both exhaust emissions and fuel consumption. However, problems associated with cycle-by-cycle variations in flame initiation and development limit the range of lean-burn operation. An experimental investigation was undertaken in order to quantify the effects of spark energy released and initial flame kernel growth on the cyclic variability of IMEP and crank angle at which 5% mass fraction was burned in a Honda VTEC-E, stratified-charge, pentroof-type, single-cylinder, optically accessed, spark-ignition engine. Simultaneous CCD images of the flame at the spark plug were acquired from two orthogonal views (one through the piston crown and one through the pentroof) on a cycle-by-cycle basis during the first 40 crank angle degrees after ignition timing, for isooctane port injection at an air to fuel ratio of 22, engine speed of 1500 RPM, 30% volumetric efficiency and 40° crank angle spark advance. The stage of the first 40° crank angle after ignition timing corresponded to the average time of 5% mass fraction burned. Image acquisition was synchronised with in-cylinder pressure, spark voltage and spark current sampling, for each of the following four spark-plug orientations: upstream, downstream and two crossflow positions of the ground electrode relative to the mean ensemble-averaged velocity vector at the spark-plug location at ignition timing. The measurements showed large cycle-by-cycle variations in flame size, shape and location. The flame size (projected enflamed areas through the two views and estimated enflamed volume) at 40° crank angle after ignition timing was found to correlate with coefficients as high as -0.96 with the crank angle of 5% mass fraction burned and 0.85 with IMEP. This degree of correlation between the flame size and the crank angle of 5% mass fraction burned or IMEP, reduced as the crank angle after ignition timing decreased, but correlations were still as large as -0.64 between the flame size at 10° crank angle after ignition timing and the crank angle of 5% mass fraction burned. The crossflow orientations produced the lowest levels of cyclic variability in IMEP, the upstream one the highest and the downstream orientation performed in between. The spark duration was found to vary from 12° to 22° crank angle on a cycle-by-cycle basis, being 19° crank angle on average, and this in conjunction with the behaviour of the flame-size correlations with the crank angle of 5% mass fraction burned suggests that the “quality” of a cycle is determined within the short time of the spark discharge. High spark energies were generally associated with short durations of the spark event and always faster than average initial flame kernel growth cycles. The correlation coefficient between spark energy and flame volume was as high as 0.76 at 10° crank angle after ignition timing, decreasing to 0.59 at 20° crank angle after ignition timing. The displacement of the flame luminous centre was not found to produce very strong correlation coefficients with the crank angle of 5% mass fraction burned (<0.55), but the correlations with the spark energy and, mainly, duration were quite strong (0.5-0.7), especially for the later stage of 20° crank angle after ignition timing rather than the earlier one of 10° crank angle after ignition timing. Although mostly associated with low spark energies, long spark durations could also produce fast cycles presumably because they provided an ignition window large enough to mask the effects of in-cylinder variations. For cycles with a given spark energy, the ones with the longest duration always produced higher than average initial flame kernel growth cycles.