This study investigated the feasibility of discriminating Chinese wines from different geographical origins using non-targeted1H NMR fingerprinting, 224 wine samples from three distinctive geographical origins (Shacheng, Changli and Changji) were measured. Principal component analysis (PCA) combined with linear discriminant analysis (LDA) using internal leave one-out cross validation were employed to establish the discrimination model of the three geographical origins. Accurate classification rates for the wine samples of Shacheng, Changli and Changji were 92%, 73%, 68%, respectively. In order to avoid the overfitting, repeated double random cross validation methods was used to verify the validity of the PCA/LDA model. This study suggested that1H NMR fingerprinting of wines contained the geographical origin feature information, the non-targeted1H NMR fingerprinting model for verifying Chinese wines based on their geographical origins could be constructed by multivariate statistical analysis technology.