In-flame soot sampling based on the thermophoresis of particles and subsequent transmission electron microscope (TEM) imaging has been conducted in a diesel engine to study size, shape and structure of soot particles within the reacting diesel jet. A direct TEM sampling is pursued, as opposed to exhaust sampling, to gain fundamental insight about the structure of soot during key formation and oxidation stages. The size and shape of soot particles aggregate structure with stretched chains of spherical-like primary particles is currently an unknown for engine soot modelling approaches. However, the in-flame sampling of soot particles in the engine poses significant challenges in order to extract meaningful data. In this paper, the engine modification to address the challenges of high-pressure sealing and avoiding interference with moving valves and piston are discussed in detail. Of particular interest is the uncertainty caused by a selection of the on-grid locations for transmission electron microscope imaging and cycle-to-cycle fluctuations of the engine combustion. Marked variations are observed in the number and projected area of soot particles depending on these variations; however, their impacts on the size of aggregates and primary particles is found to be minor. Also, the morphology of soot particles appears not to be sensitive to the exposure duration of the grid to the sooting flame; however, the duration is ultimately limited by soot over-loading. Two different injection pressures are selected to test the usefulness of the in-flame soot sampling in a diesel engine and the results show that a decrease in the size of soot aggregates and primary particles with increasing injection pressure exceeds the uncertainty.