Because insurgent group formation typically occurs in secrecy and in poorly monitored areas, the empirical record on conflicts’ start is spare and systematically omits rebels who fail before committing substantial violence. This article argues that this presents a fundamental challenge for the study of conflict onset and demonstrates the theoretical and empirical problems it causes in studying a controversial relationship: how ethnicity influences armed conflicts’ start. Unusual evidence on all armed groups that formed in Uganda since 1986 indicates that ethnic mobilization was unimportant to the initial formation of rebel groups—but mattered after nascent groups had already formed. Contrasting evidence from Uganda with a prominent argument that ethnic marginalization induces rebellion shows why lack of evidence about how insurgencies begin can lead to broader inferential pitfalls.