This paper presents a rapid and non-destructive approach to distinguishing mineral, purified, and tap water using near-infrared (NIR) transmission spectroscopy and chemometrics in the 900-1700 nm spectral region. Following z-score standardisation, Principal Component Analysis (PCA) was implemented to investigate spectral variance. The first principal component accounted for 67.78% of the overall variance, showing a significant discriminative structure in the dataset. Fisher discriminant analysis identified crucial wavelengths at 1155 nm, 1159 nm, and 1244 nm as the most informative spectral characteristics corresponding to second-overtone C-H and combination-band O-H absorption. Stratified 10-fold cross-validation was used to evaluate three supervised classification algorithms: k-Nearest Neighbour (KNN), Support Vector Machine with radial basis function kernel (SVM-RBF), and Linear Discriminant Analysis (LDA). With only one misclassification out of 120 samples, the confusion matrix revealed that the SVM-RBF and LDA classifiers had a typical classification accuracy of 99.2%. These results reveal that NIR spectroscopy offers an immediate, reagent-free, and highly sensitive technique for water source verification and quality monitoring when combined with proper pre-processing and chemometrics.