The authors model the duration of large-scale, violent civil conflicts, applying hazard functions to a comprehensive data set on such conflicts for the period 1960-99. They find that the duration of conflicts is determined by a substantially different set of variables than those that determine their initiation. The duration of conflict increases substantially if the society is composed of a few large ethnic groups, if there is extensive forest cover, and if the conflict has commenced since 1980. None of these factors affects the initiation of conflict. The authors also find that neither the duration nor the initiation of conflict is affected by initial inequality or political repression. This finding is consistent with the hypothesis that rebellions are initiated where they are viable during conflict, regardless of the prospects of attaining post-conflict goals, and that they persist unless circumstances change.