Many variables such as total soluble sugars, total titratable acidity, nitrogen and phenolic compounds balance, contribute to describe grape quality. They vary strongly with genetic (cultivars) and environmental (climate, soil, and cultural practices together referred to as “terroir”, and vintage) factors. The aim of this work was to determine changes in metabolite fingerprints of grape berry skins of ‘Merlot noir’, ‘Carbernet franc’ or ‘Cabernet Sauvignon’ cultivars harvested in 2002, 2003 and 2004 from five geographical locations in Bordeaux (south-west of France) to better understand the factors influencing the grape berry composition. 1H NMR spectroscopy was used to determine metabolic profiles of skin extracts of berries harvested at maturity. Before analysis by multivariate statistical methods, spectral data were reduced (sum of intensities over 0.04 ppm spectral domains) and normalized to generate 183 variables describing the entire spectra. Principal component analysis (PCA) and partial least squares (PLS) chemometric methods were applied to describe sample variability and show clustering of samples. These chemometric methods gave a good separation of samples according to vintages. PLS allowed us to pinpoint spectral domains corresponding to metabolites contributing to the differences between vintages, including sugars (glucose, fructose and sucrose), organic acids (tartaric, malic, citric and succinic acids), and amino acids (proline, arginine, GABA, valine, alanine, leucine and isoleucine). Conversely, soils classified according to their capacity to provide water to the vines, could not be discriminated by 1H NMR metabolic fingerprinting of the berries. The vintage effect on grape metabolic profiles prevailed over the soil effect. This emphasizes the fact that a typical vineyard is defined by its most frequent climatic traits such as the seasonal sum of temperatures and water balance.