We analyse the structure of food webs from a reliability theoretical point
of view. The reliability of the network flow is the probability that source
s are connected to sinks. With a fixed number of species and connectance, r
eliability depends on the pattern of links. We construct all of the possibl
e sink web graphs with given numbers of species and links and calculate the
reliability of the network flow for each hypothetical sink web. Patterns f
avourable for reliable flows are characterized. Having compared our results
with real web data, we conclude that the structure of natural webs contain
s mostly such patterns. Thus, the reliability of the network flow seems to
be an important factor governing food web structure. Our results show that
short chains, a high level of omnivory, a high prey-predator ratio and a lo
w number of specialists are favourable for reliable network flows. These co
nclusions support the reticulate nature of food webs and the rethinking of
the traditional, linearized view.