We examine a major source of heterogeneity across cases in the Correlates of War Militarized Interstate Dispute Dataset, 1816–2001, and demonstrate that this variation across cases biases most analyses of conflict. Disputes are coded using two logics—the familiar state-to-state militarized action represents one case while the second relies on sponsor governments to protest state targeting of private citizens. We show that the latter introduces additional measurement bias and does not match well the original conceptualization of what constituted a dispute. The protest-dependent cases are caused by different processes, and omitting them from analyses provides truer estimates of the effects of most conflict predictors. We find that previous controls for heterogeneity in the dispute data—such as using fatal militarized interstate disputes only—substantially underestimates the dangerous effects of contiguity and the pacifying effects of regime similarity. We also show that governments are seldom willing to risk militarized conflict for private citizens during these unique cases. We provide a list of the protest-dependent cases for future conflict analyses.