Previous numerical studies of community assembly have found (1) that i
nvasion resistance increases with time, and (2) that different assembl
y sequences typically result in different community endpoints. The alg
orithm used in these studies involved sequential introductions of spec
ies coupled with tests for existence of feasible equilibria with local
asymptotic stability. In this paper the algorithm is tested against a
method based on numerical integration. We show that the algorithm con
tains serious technical flaws. The algorithm gives the incorrect outco
me in many iterations, and correct assembly sequences diverge rapidly
from those it predicts. In the light of this, we reassess the earlier
work and suggest that results (1) and (2) above should be treated with
caution. A more reliable method of community assembly is clearly need
ed. Numerical integration, although useful for investigating the outco
me over small numbers of iterations;is too slow to deal with large, hi
ghly replicated, assembly sequences. We suggest that an alternative, b
ased on a criterion of coexistence known as permanence, would be appro
priate, as this is relatively fast and reliable.