Tropical forests of the Amazon Basin are being rapidly converted to agricultural land uses and fallow land, resulting in accelerating rates of forest loss in one of the world's most biodiverse ecoregions. This process has been extensively described and modelled, but as yet there has been no formal test of how the spatial patterns of deforested and fragmented areas change with the spatial scale of forest clearings. It was hypothesised that different land-use practices are driving small and large clearings, with small-scale cultivators often creating small, irregularly shaped clearings and large-scale ranchers and soy farmers creating larger, more regular-shaped clearings. To quantitatively test this hypothesis, Mandelbrot's theory of fractals was applied to deforested areas in the Brazilian Amazon to test for scale-invariance in deforestation patterns. The spatial pattern of deforestation differed between small and large clearings, with the former creating more complex landscapes and with a threshold occurring at c. 1200 ha in area. As a consequence, the sizes and shapes of forest clearings, and hence the relative vulnerability of the remaining forest to edge, area and isolation effects, may differ systematically between landscapes with different deforestation drivers. Further tests of this hypothesis are needed to assess its efficacy in other tropical landscapes and geographical locations.