Simple neutral percolation models of ecological landscapes predict a c
ritical threshold of landscape connectivity at ca 60% coverage of cell
s that are permeable to movement. Using such models as an inspiration,
we conducted a field experiment in which we employed an experimental
model system of tenebrionid beetles (Eleodes obsoleta) moving through
mosaics of grass and bare ground in which grass coverage was systemati
cally varied from 0% to 80% in a random pattern. Three parameters of b
eetle movement pathways (mean step length, mean vector length, and net
displacement rare) all showed a sharp reduction between 0% and 20% gr
ass cover and no differences over further 20% coverage increments; the
mean Fractal dimension of pathways (which could not be derived for th
e 0% coverage treatment) also did not differ with increasing grass cov
erage from 20% to 80%. The proportion of time spent by beetles in gras
s patches did not increase with increasing grass coverage, although in
dividuals spent significantly more time steps in grass in the 20% gras
s-cover treatment than would be expected by chance. The distance moved
per time step was greater when beetles moved over bare ground than in
grass, but the mean step length while on grass was significantly lowe
r when grass constituted only 20% of the experimental landscape than w
hen more grass was present, perhaps because beetles stopped more frequ
ently on grass when little of it was present. The threshold in beetle
movement through the experimental landscapes differed markedly from th
at predicted by simple neutral percolation models, probably because, i
n contrast to the models, both cover types were permeable to beetle mo
vements and individuals moved nonrandomly. The results of this experim
ent indicate that definitions of landscape connectivity depend on both
the spatial pattern of the landscape and how individuals move within
and among patches. Moreover, it appears that movement patterns within
a particular patch may be contingent on the characteristics of the sur
rounding landscape. As a consequence, a land-cover map alone may not p
redict whether, to a particular kind of organism, a landscape is fragm
ented or connected, and the spatial distribution of individuals in a p
opulation may not map closely onto the distribution of suitable habita
t patches. This finding suggests that assessments of metapopulation st
ructure must consider both overall landscape patterns and the nonlinea
r responses of organisms to such patterns.