Resummation of large infrared logarithms in perturbation theory can, in certain
circumstances, enhance the sensitivity to small gluon momenta and introduce
spurious nonperturbative contributions. In particular, different procedures --
equivalent in perturbation theory -- to organize this resummation can differ by
$1/Q$ power corrections. The question arises whether one can formulate
resummation procedures that are explicitly consistent with the infrared behaviour of
finite-order Feynman diagrams. We explain how this problem can be treated and
resolved in Drell-Yan (lepton pair) production and briefly discuss more
complicated cases, such as top quark production and event shape variables in the
$e^+e^-$ annihilation.