Roselle (Hibiscus sabdariffa) is a functional food with potential health benefits, consumed either as hot or cold beverage. To ensure quality control of its various products, accurate measurement of active metabolites is warranted. Herein, we propose a combination of ultra-high performance liquid chromatography-mass spectrometry (UHPLC-MS) and nuclear magnetic resonance (NMR) analytical platforms for the untargeted characterization of metabolites in two roselle cultivars, Aswan and Sudan-1. The analyses revealed 33 metabolites, including sugars, flavonoids, anthocyanins, phenolic and aliphatic organic acids. Their relative contents in cultivars were assessed via principle component analysis (PCA) and orthogonal projection to latent structures analysis (OPLS). Impact of the different extraction methods (decoction, infusion and maceration) was compared by quantitative H-1 NMR spectroscopy, revealing cold maceration to be optimal for preserving anthocyanins, whereas infusion was more suited for recovering organic acids. The metabolite pattern revealed by the different extraction methods was found in good correlation for their ability to inhibit alpha-glucosidase enzyme.