Objective: Weight variation during therapy has been described as a useful marker to predict TB treatment outcome. No previous study has used longitudinal analysis to corroborate this finding. The goal of this study was to evaluate change and trends of patients' bodyweight over time depending on TB treatment outcome. Methods and Findings: A retrospective cohort study with all TB cases diagnosed from 2000 to 2006 was carried out. Information from 5 public tuberculosis treatment facilities at Pampas de San Juan de Miraflores, Lima, Peru was analyzed. Poor outcome was defined as failure or death during TB therapy, and compared to good outcome defined as cured. Longitudinal analysis with a pre-specified marginal model was fitted using Generalized Estimating Equations to compare weight trends for patients with good and poor outcome adjusting for potential confounders. A total of 460 patients (55.4% males, mean age: 31.6 years) were included in the analysis: 42 (9.1%) had a poor outcome (17 failed and 25 died). Weight at baseline was not different comparing outcome groups (p = 0.17). After adjusting for age, gender, type of TB, scheme of treatment, HIV status and sputum variation during follow-up, after the first month of treatment, patients with good outcome gained, on average, almost 1 kg compared to their baseline weight (p<0.001), whereas those with poor outcome lost 1 kg (p = 0.003). Similarly, after 4 months, a patient with good outcome increased 3 kg on average (p<0.001), while those with poor outcome only gained 0.2 kg (p = 0.02). Conclusions: Weight variation during tuberculosis therapy follow-up can predict treatment outcome. Patients losing weight during TB treatment, especially in the first month, should be more closely followed as they are at risk of failure or death.