A test-retest study was carried out to assess the reproducibility of a semi-quantitative food frequency questionnaire (SFFQ) designed for adolescents. The study group included 108 individuals. The mean interval between the two interviews was 20 days. Three days of food records from 430 adolescents were used to define the items and portions in the SFFQ. Intraclass correlation coefficients (ICC) were estimated to assess the agreement in food item reports. The agreement for energy and nutrient intake was assessed by Pearson's correlation coefficients, weighted kappa (agreement in quartile classification), and the Bland-Altman method. Data for nutrients were adjusted according to energy intake. Mean ICC was 0.54. The categorization in the same and in the adjacent quartile for data adjusted for energy intake ranged from 72% to 86% (mean = 80%). According to the Bland-Altman method, mean agreements ranged from 77% to 104% and were independent of the intake variation, except for vitamin A. Pearson's correlation coefficients ranged from 0.40 to 0.61 (mean = 0.52). The SFFQ showed acceptable reliability to assess food, energy, and nutrient intake by adolescents in the study area.