Food quality and safety are closely related to the geographical origin of food. This study combined mineral element analysis and multivariate statistical analysis to discriminate the origins of 148 blueberry samples from three regions of China. The concentrations of K, Ca, Mg, Na, Fe, Cu, Mn, B, P, and Zn were determined inductively coupled plasma-atomic emission spectroscopy (ICP-AES). Variance analysis (ANOVA), Duncan’s multiple-comparison test, and principal component analysis (PCA) were used to compare the element concentrations, and statistically significant differences were found among samples from different regions. Linear discriminant analysis (LDA), decision tree (DT), multilayer perceptron neural network (MLP-NN), and support vector machine (SVM) were utilized to build models for blueberry authentication. The results showed that the average concentrations of the minerals were in the order of K > P > Ca > Mg > Na > Mn > Fe > Zn > B > Cu, and the levels of K, Ca, Mg, Fe, Cu, Mn, B, P, and Zn were significantly different among regions by ANOVA and Duncan’s multiple-comparison test. The study indicates that LDA, DT, MLP-NN, and SVM chemometric tools have the potential to discriminate the origin of blueberries. The results revealed that the MLP-NN and SVM models were more discriminative than the other two mathematical methods. The MLP-NN yielded an average discrimination rate of 92.7% for the training set and 94.7% for the test set, and the SVM with linear kernel function (SVM-lin) obtained an average identification rate of 91.8% for the training set and 94.7% for the test set. The order of successful identification rates was as follows: MLP-NN > SVM-lin > DT > LDA. This study can serve as a reference to identify the origin of blueberries and perform quality assurance for the fruit.