A statistical regularity has been documented in several empirical studies: More unequal countries tend to show a higher degree of social disorder. Peru is a country with recurrent political instability and social disorder that also displays a pronounced degree of inequality. What is the role played by ethnicity in this relationship? In this paper we propose a new way of defining ethnic groups in Peru based on Peruvian geography and history, which corrects the standard view based on language differences alone. With this new definition we contrast the empirical hypothesis of three theoretical frameworks relating inter-group inequality and social disorder. We present empirical estimates of interethnic inequalities on human capital, labour market, and income. The econometric results show that the educational level of people depends upon ethnicity; moreover, there is exclusion, not discrimination, in the access to high skilled positions. We evaluate the roles of different social actors in the reduction of inequality. Although the indigenous populations have experienced significant gains in absolute terms, they have not experienced gains in relative terms. Therefore, horizontal inequalities in Peru are significant and persistent, and contribute largely to overall inequality. The role of horizontal inequalities in the instability of Peru seems to be important, but as a latent factor. Horizontal inequalities do contribute to the social disorder in Peru, but not much in a direct way. Ethnic conflict is not the prime mover of social disorder. This apparent paradox is explained by the fact that Peru is a multiethnic and hierarchical society, where the indigenous populations are second rate citizens. In sum, in explaining inequality in Peru, ethnicity matters. These empirical results are consistent with the predictions of Sigma Theory (Figueroam 2003) and with some of the predictions of Horizontal Inequality Theory (Stewart 2001), but inconsistent with Neoclassical Theory, even when social heterogeneity is introduced in its analysis (Becker and Murphy 2000).