The untargeted metabolomics of Newhall navel oranges from three areas in China-Ganzhou, Fengjie, and Zigui-with geographical indication (GI) was measured using LC-MS/MS. Orthogonal partial least squares discriminant analysis was performed for sample classification and important metabolite identification. This approach identified the best markers of the geographical origin able to discriminate Fengjie, Ganzhou, and Zigui orange samples. For peeled samples, 2-isopropylmalic acid, succinic acid, citric acid, L-aspartic acid, L-glutamic γ-semialdehyde, D-β-phenylalanine, hesperetin, hydrocinnamic acid, 4-hydroxycinnamic acid, and dehydroascorbate were the markers used to discriminate the geographical origin. All these markers were overexpressed in the peeled samples from the Zigui area, followed by the Ganzhou area. As for unpeeled samples, L-glutamic γ-semialdehyde, isovitexin 2'-O-β-D-glucoside, 2-isopropylmalic acid, isovitexin, diosmetin, trans-2-hydroxycinnamate and trans-cinnamate, L-aspartic acid, hydrocinnamic acid, and β-carotene were used to discriminate their origin. The first seven markers in Zigui-planted whole samples showed the highest levels, and the last three markers were richest in Ganzhou-planted samples. According to the variation in the markers for discriminating the origins of the peeled or unpeeled Newhall navel oranges with GI and the highest value of titratable acidity in those from Zigui, the samples planted in Ganzhou have the best balance between taste and nutrition. This work confirms that the approach of untargeted metabolomics combined with OPLS-DA is an effective way for origin tracing and overall quality evaluation.