Habitat fragmentation is a potentially critical factor in determining popul
ation persistence. In this paper, we explore the effect of fragmentation wh
en the fragmentation follows a fractal pattern. The habitat is divided into
patches, each of which is suitable or unsuitable. Suitable patches are eit
her occupied or unoccupied, and change state depending on rates of coloniza
tion and local extinction. We compare the behaviour of two models: a spatia
lly implicit patch-occupancy (PO) model and a spatially explicit cellular a
utomaton (CA) model. The PO model has two fixed points: extinction, and a s
table equilibrium with a fixed proportion of occupied patches. Global extin
ction results when habitat destruction reduces the proportion of suitable p
atches below a critical threshold. The PO model successfully recreates the
extinction patterns found in other models. We translated the PO model into
a stochastic cellular automaton. Fractal arrangements of suitable and unsui
table patches were used to simulate habitat fragmentation. We found that: (
i) a population on a fractal landscape can tolerate more habitat destructio
n than predicted by the patch-occupancy model, and (ii) the extinction thre
shold decreases as the fractal dimension of the landscape decreases. These
effects cannot be seen in spatially implicit models. Landscape structure pl
ays a vital role in mediating the effects of habitat fragmentation on persi
stence.