This article revisits the arms race to war relationship with the hope of resolving a lingering debate in international relations over the effects of arms races. Previous empirical studies in this area suffered from a possible selection effect, rendering them unable to differentiate between the escalatory and deterrent effects of arms races. Specifically, earlier quantitative investigations were unable to test deterrence hypotheses, because the unit of analysis (dispute) presupposed that deterrence had already failed in preventing dispute onset. In order to take the possibility of deterrence seriously, a dataset is constructed that identifies arms races independently of dispute occurrence. This article improves on previous studies in that a measure of interdependent arming exogenous to dispute initiation allows for a test of whether arms races actually deter the onset of militarized disputes or contribute to dispute escalation. Both the deterrence and escalation hypotheses are tested using a sample of 'strategic rivals' from 1816 to 1993. The analyses reveal that arms races increase the likelihood of disputes and war. Furthermore, to account for the possibility that the arms race to war relationship may be spurious to dyadic hostilities accounting for both arms races and war, a selection model is employed that differentiates between dispute and war processes. This indicates that arms races do not contribute to deterrence and are instead associated with both disputes and war.