This chapter investigates the application of cluster analysis to the characterization of dynamic positron emission tomography (PET) images. Cluster analysis is one of several data-led techniques that are of potential value in the analysis of PET data. This technique can be used to partition the large number of pixel time–activity curves, obtained from a dynamic scan into a smaller number of clusters. Effectively each cluster represents a characteristic “shape” of a time–activity curve. The likelihood of any pixel vector belonging to a given cluster can be computed. This then enables a corresponding partition or segmentation of the dynamic image set into images of the spatial distribution of each of the clusters. Because the cluster means are derived from many pixels, they exhibit a much improved signal-to-noise ratio. This partitioning requires no knowledge of the plasma input function. However, because the cluster means are in the same space as the original data, further model analysis and characterization of the cluster means relative to an input function are possible.