In QCD, perturbation theory is an indispensable tool but is not an ultimate method in the sense that accuracies of perturbative predictions are limited. In particular, the so-called renormalon problem is known that perturbative expansions of observables give divergent series and cause inevitable uncertainties. This problem is becoming more serious from phenomenological aspects in recent years. I first give a review of the renormalon problem. The contents include causes of the divergence of perturbative series and its implications to nonperturbative physics. Subsequently, I discuss a possible direction to overcome the renormalon problem. For this purpose, one has to calculate perturbative contributions compatibly with the operator product expansion so that nonperturbative effects can eventually be included.