In this article, we re-examine the statistical evidence for the democratic peace at the dyadic level. We also investigate the seeming paradox that democracies are engaged in war as often as autocracies at the nation level. From the extensive literature on democracy and peace we have selected as our point of departure two influential contributions by Stuart Bremer and Zeev Maoz & Bruce Russett, which both analyze the relation between democracy and peace at the dyadic level. Several problematic aspects of their analyses are addressed; in particular problems with dependence between observational units caused by continuing war and peace, and by diffusion effects. We show that the increasing number of countries in the international system causes their assumption of a stationary probability of war at the dyadic level to be violated. It is argued that these problems cannot be solved adequately within the traditional dyad-year framework. Instead it is proposed to model observations on the interstate dyad as a process in continuous time, using Cox regression. An extensive model is developed that controls for contiguity, power status, alliances, stability, diffusion of war, and recurrence effects. A concept of relevance is introduced to account for the dependence of the dyadic probability of war on the size of the international system. The democratic peace is supported in our basic model. In a refined model, we find that democracies’ tendency to join each other in wars is much more marked than their avoidance of mutual fighting. This explains why democracies are as war-prone as autocracies.