With the increasing interest in hydrogen propulsion in the automotive sector, research regarding optimisation of carbon fibre composite pressure vessels (CPVs) has gained considerable momentum in past few years. CPVs are produced by filament winding of continuous unidirectional fibres over a polymer liner (Type IV). Due to high in-service pressure, it is desired that the CPV has minimal defects, but the filament winding process inevitably introduces microstructural defects such as voids, fibre misalignment and local fibre volume fraction variation. To be able to optimise the CPV performance, it is vital to understand the factors affecting the microstructure. X-ray computed tomography (XCT) is used extensively in field of composites for microstructural analysis [1]. However, the varying scale of the microstructural defects in CPV necessitates a combination of high and low-resolution imaging as well as optimisation of the number of scan samples for a representative analysis of the full CPV. This study proposes a reliable approach for acquiring a representative charaterisation of CPV microstructure with minimum number of scans. We determine the scanning strategy for a CPV, including both the scale (scan resolution) and the scope (number and locations of scans) of the analysis (Figure 1). A sensitivity analysis is performed for the determination of the size of studied features and the corresponding optimum resolutions and voxel sizes. A voxel size of 1 µm is found to be ideal for fibre related features, especially local fibre volume fraction, while 2.5 µm is found to be ideal for the majority of voids in hoop layers. Another issue is the selection of the number and positions where samples should be taken for a statistically correct representation of spatial variation on the vessel. The statistical analysis reveals that in the circumferential direction, a minimum of five well distributed samples are sufficient (subject to selected resolution). The results acquired from this analysis can be used for optimising the predictive models as well as to understand and improve the mechanical performance of CPVs.