The spatial pattern of built-up areas within a NUTS-1 European region (Wallonia in Belgium) is analysed using fractal indices. Methodologically, this paper illustrates the usefulness of fractal indices in measuring built-up morphologies, and also shows that clustering techniques have to be adapted for the non-Euclidean nature of the fractal measurements. An expectation maximisation algorithm (EM) combined with a Bayesian information criterion (BIC) is used. Empirically, we show that fractal indices partition the region into sub-areas that do not correspond to “natural landscapes” but result from the history of urbanisation. Urban sprawl seems to affect most communes, even the remotest villages: traditional (compact, ribbon, etc.) villages are transformed into more complex and heterogeneous shapes. These indices seem to be useful for characterising and understanding the built landscapes, as well as for modelling and planning urban realities.