Stringent particulate emission regulations are applied to spark-ignition direct-injection (SIDI) engines, calling for a significant in-cylinder reduction of soot particles. To enhance fundamental knowledge of the soot formation and oxidation process inside the cylinder of the engine, a new in-flame particle sampling system has been developed and implemented in a working optical SIDI engine with a side-mounted, wall-guided injection system. Using the sampling probes installed on the piston top, the soot particles are directly sampled from the petrol flame for detailed analysis of particle size distribution, structure and shape. At the probe tip, a transmission electron microscope (TEM) grid is stored for the soot collection via thermophoresis, which is imaged and post-processed for statistical analysis. Simultaneously, the flame development was recorded using two high-speed cameras to evidence the direct exposure of the sampling grids to the soot-laden diffusion flames and pool fires. The focus of the present study is the uncertainty analysis of this newly developed technique through variation of the number of injection cycles, cyclic variations, and sampling locations at fixed fuel injection and firing conditions. From the engine runs of 3, 5 and 7 injection cycles, it was found that the number of sampled soot aggregates increases with increasing injection cycles but the soot morphology does not change significantly. However, the cyclic variations make a significant impact such that the size of soot aggregates increases with higher peak in-cylinder pressure and earlier combustion phasing, which was observed from 6 different engine runs at fixed 5 injection cycle tests. The sampling experiment was also performed with the probes installed at four different locations across the piston top, which showed significant variations in morphology such that soot primary particles and aggregates become larger due to longer soot residence time within the wall wetting-induced pool fire.