Background. Globally, >30 000 children fall sick with multidrug-resistant (MDR) tuberculosis every year. Without robust pediatric data, clinical management follows international guidelines that are based on studies in adults and expert opinion. We aimed to identify baseline predictors of death, treatment failure, and loss to follow-up among children with MDR tuberculosis disease treated with regimens tailored to their drug susceptibility test (DST) result or to the DST result of a source case. Methods. This retrospective cohort study included all children ≤15 years old with confirmed and probable MDR tuberculosis disease who began tailored regimens in Lima, Peru, between 2005 and 2009. Using logistic regression, we examined associations between baseline patient and treatment characteristics and (1) death or treatment failure and (2) loss to follow-up. Results. Two hundred eleven of 232 (90.9%) children had known treatment outcomes, of whom 163 (77.2%) achieved cure or probable cure, 29 (13.7%) were lost to follow-up, 10 (4.7%) experienced treatment failure, and 9 (4.3%) died. Independent baseline predictors of death or treatment failure were the presence of severe disease (adjusted odds ratio aOR, 4.96; 95% confidence interval CI, 1.61–15.26) and z score ≤−1 (aOR, 3.39; 95% CI, 1.20–9.54). We did not identify any independent predictors of loss to follow-up. Conclusions. High cure rates can be achieved in children with MDR tuberculosis using tailored regimens containing second-line drugs. However, children faced significantly higher risk of death or treatment failure if they had severe disease or were underweight. These findings highlight the need for early interventions that can improve treatment outcomes for children with MDR tuberculosis.