This dissertation explains why security sector reform (or SSR), a widely adopted measure in post-conflict contexts, sometimes leads to very high levels of criminal violence. In contrast to prevailing explanations that emphasize the role of SSR in creating the conditions for peace, I focus on the police--the state institution usually in charge of enforcing the law domestically--and argue that conditional on the type of reform adopted, SSR can create conditions for increased violence. In the aftermath of armed conflict, the international community intervenes to reform the security apparatus with specific prescriptions. Such prescriptions usually focus on two key areas: accountability--the generation of information and imposition of costs on behavior that departs from certain rules, based on the assumption that legal consequences will deter police from excessive violence--and militarization--understood as changes in weaponry and training that will deter others from engaging in violence.\indent These prescriptions lose sight of two fundamental features of post-conflict settings: 1) the presence of pre--existing values regarding the use of violence; 2) the logic of the strategic interactions between the police and violent actors. I theorize that reforms that do not reduce militarization tend to produce police forces that are more prone to use violence. I theorize further that accountability mechanisms reduce police propensity for violence only under conditions of low militarization. I conceptualize demilitarization as a change in the logic of the use of violence (from the elimination of an enemy combatant, to the protection of a citizen with rights). Accountability refers to the imposition of constraints (disciplinary actions and oversight) on the use of violence by the police and other violations of its code of ethics such as corruption.\indent Reforms operate to increase or decrease police propensity to use force; this behavioral change has an effect on criminal violence. Prior to the end of conflict, police and criminals interact frequently. These repeated interactions lead criminals to form expectations about police behavior towards them. Reforms affect such expectations. As police propensity for violence increases, it affects criminal groups' ability to anticipate police behavior, creating uncertainty. This is because police's propensity for violence is manifested selectively. Criminal groups cannot anticipate whether there will be violence towards them, or to what extent. Under conditions of uncertainty, the rational strategy is to respond with a combative behavior. In anticipation of the possibility of police violence, criminal groups will tend to arm up and recruit more members. The condition of uncertainty generates spirals of violence.\indent I rely on a mixed-methods strategy to test this argument. I built an original dataset containing detailed information on police reforms implemented in all countries that transitioned out of a conflict between the end of the Cold War and 2015. I conceptualize reforms as concrete changes to the structure and functions of the police, regardless of the intended outcomes of the reform process. To construct this dataset, I operationalized the dimensions of militarization and accountability in a series of sub-dimensions based on the theoretical argument. The dataset triangulates multiple sources of information--case studies by non-governmental organizations, police encyclopedias, police websites or national documents, US State Department and Amnesty International Human Rights Country Reports--and it improves upon others by disaggregating reforms into separate components, which allows for additional variation in the spectrum of reforms. I use this database to conduct a cross-national analysis of the effect of reforms on police and criminal violence. I then explore the causal mechanisms underpinning my argument through within--case comparative analysis of the South African experience following extensive reforms implemented in 1994. \indent Results from quantitative analyses show that countries that implement reforms that increase militarization tend to experience greater increases in violence post--conflict than those that do not. The analyses also suggest that accountability mechanisms, although relevant to reduce police propensity for violence, have a more limited effect, at lower levels of militarization. At high levels of militarization in particular, accountability has no effect on police violence, suggesting that it operates to reinforce militarization. I also show that structural factors such as inequality, poverty, and weak institutions are not sufficient to explain why violence is reduced in some post-conflict contexts but not others. \indent To delve deeper into how these processes occur, I conducted extensive fieldwork in South Africa. I interviewed government officials, members of the police, academics, and members of think-tanks and non-governmental organizations. I also undertook a review of twenty years of parliamentary debates on police reform and the construction of a database of newspaper articles to understand the patterns of confrontation between police and criminal groups between 1989 and 2015. The case study provides rich evidence of the mechanisms driving these changes. Through a reconstruction of the processes following the reforms, I show that reductions in militarization happened by implementing changes in composition and organization of the police service under the Mandela administration. The case also shows that the accountability structures implemented increased the cost of violations and police use of violence decreased when militarization was low. Conversely, as militarization increased in the 2000s, additional organizational changes led to an increase in police violence, even in the presence of accountability, which escalated criminal violence. \indent My research shows that security reforms, particularly of the police, are a key element in efforts to reduce post--conflict criminal violence because they can exacerbate the conditions that sustain it. More importantly, I show that violence is not a transitory problem some states face on their way to democratization; rather, it can become a permanent feature of daily social interactions. The creation of an effective rule of law depends in part on its enforcement, which falls within the purview of the police. The abuse of force as a mechanism to implement the rule of law selectively, by treating individuals or groups as enemies, creates conditions for increased violence, which is detrimental to security, peace, and ultimately human development. Post--conflict criminal violence is not a phenomenon of the post-transition period, rather, its ebbs and flows result from the legacies of complex dynamics between violent actors, the state, and ordinary citizens that developed during conflict. Importantly, many of the lessons from my project extend beyond post--conflict cases, and can apply to democratic contexts where militarization of the police is taking place.