Why are some wars longer than others? Offense-defense theorists have assumed that wars are shorter and it is more difficult for states to create security when military technology favors the offense and attacking is easier than defending. This study argues that this assumption is wrong on both accounts. First, statistical hazards models show that various measures of the offense-defense balance are not associated with war duration in the state system between 1817 and 1992. Second, if wars are not shorter when offense is dominant, there should not be more rational incentives for decision-makers to start wars when offense has the advantage compared to when defending is relatively easier. The study develops the bounded learning theory to better understand the causes of war duration and tests it with four case studies. The theory argues that offense dominance is often so limited that it fails to guarantee that theenemy is swiftly run over, other things being equal. As a swift victory does not materialize with the help of offensive military technology and tactics, the process of finding a mutually acceptable negotiated solution to the war easily becomes long because of asymmetric information about expected offensive capacity and expansive ideology (asymmetric causal beliefs and offensive stakes). These variables make it unlikely that the combatants would quickly adjust their war aims so that they reflect their actual battlefield performance. As the combatants do not learn from the battlefield events, they cannot agree on their expected relative strength, which is necessary from a rational choice perspective for finding common ground for a peace treaty. Offense-defense theory specifies defense dominance as the prerequisite for more peaceful interstate relations. On the other hand, defense dominance is difficult to create as the realist self-help logic drives many states to continue developing offensive weapons technology. Furthermore, by assuming that wars are shorter during offense dominant eras, the offense-defense theory creates false incentives for aggression. It is common sense that the risk of war does not increase when defense is dominant. However, even when attacking is relatively easier than defending, incentives for aggression should not increase on average if wars are not particularly short. Thus the study is a contribution to defensive realism.