A novel technique for identifying and characterising clusters of particles from measurements within a densely seeded two-phase flow is reported. This technique involves the smoothing of normalised instantaneous planar images of particle concentration followed by the application of a robust and unambiguous dynamic threshold to identify particle clusters. Also reported is a method to extract quantitative cluster data including cluster length, width and number of branches. The method employs an algorithm to morphologically skeletonize images of clusters, and subsequently, prune skeleton branches to select those which most strongly represent the shape of the cluster. Together, these techniques have been shown to identify and characterise two-dimensional slices of three-dimensional particle clusters of complex shapes, including those that are bent, wrinkled and branched, with an uncertainty of ≈4% relative to the manually determined values. This method was applied to planar measurements of particles in a heavily seeded turbulent jet with an exit Stokes number of SkD=1.4 and Reynolds number of ReD=10,000, based on the pipe diameter, D. The results show that particle clusters are already present at the exit plane and have a characteristic width that is narrowly distributed around an average value of ≈0.17D. This implies that particle clusters are generated inside the pipe at this preferred length scale. The results also show that the average cluster length at the pipe exit is ≈1.0D, which, together with the observation that the clusters are oriented at oblique angles to the axis of the jet, suggests that the length of these clusters within the pipe is limited by the pipe diameter. The aspect ratio of the cluster slices was found to be typically AR≈6−7, consistent with the observations that the clusters form long, thin, filament-like structures.