This study aims to evaluate the potential of (1)H NMR spectroscopy, combined with multivariate statistics, for discriminating between tumour and non-involved (control) pulmonary parenchyma and for providing biochemical information on different histological types. Paired tissue samples from 24 primary lung tumours were directly analysed by high-resolution magic angle spinning (HRMAS) (1)H NMR spectroscopy (500 MHz), and their spectral profiles subjected to principal component analysis (PCA) and partial least squares regression discriminant analysis (PLS-DA). Tumour and adjacent control parenchyma were clearly discriminated in the PLS-DA model with a high level of sensitivity (95% of tumour samples correctly classified) and 100% specificity (no false positives). The metabolites giving rise to this separation were mainly lactate, glycerophosphocholine, phosphocholine, taurine, reduced glutathione and uridine di-phosphate (elevated in tumours) and glucose, phosphoethanolamine, acetate, lysine, methionine, glycine, myo- and scyllo-inositol (reduced in tumours compared to control tissues). Furthermore, PLS-DA of a sub-set of tumour samples allowed adenocarcinomas to be discriminated from carcinoid tumours and epidermoid carcinomas, highlighting differences in metabolite levels between these histological types, and therefore revealing valuable knowledge on the biochemistry of different types of bronchial-pulmonary carcinomas.