Nutrient deficiencies on grapevines could affect the fruit yield and quality, which is a major concern in vineyards. Nutrient deficiencies may be recognizable by foliar symptoms that vary by mineral nutrient and stress severity, but it is too late to manage when visible deficiency symptoms become apparent. The nutrient analysis in the laboratory is the way to get an accurate result, but it is time and cost-intensive. The differences in leaf nutrient levels also alter spectral characteristics outside of the electromagnetic spectrum’s visible range, even before visible deficiency symptoms become apparent. The previous studies regarding remote sensing in vineyards are restricted to the UAVs, which captures an average reflectance observed across a vineyard block. These approaches do not reflect spatial variability within and between canopies. In this study, hyperspectral images acquired from a ground-based on-the-go utility vehicle were analyzed to evaluate the applicability of hyperspectral data in classifying nutrient deficiency into different classes. Two Nitrogen trials were conducted in two vineyards with two varieties, Sauvignon blanc and Syrah. One Potassium trial is conducted in one Chardonnay vineyard. There were four treatment levels in the Nitrogen experiment while three treatment levels in the Potassium experiment. During Veraison, hyperspectral image data (400-1000nm) was collected in both sun-exposed and diffused light conditions. The leaf samples were collected the same day from the same vines, then Nitrogen and Potassium content were analyzed in the laboratory. Principal component analysis and successive projection algorithms were applied to select key wavelengths related to nutrient content. Common wavelengths were identified. Partial least squares discriminant analysis was performed to identify different nutrient deficiency classes with the selected common wavelengths. The result shows the models could classify nutrient deficiency into three levels for both Nitrogen and Potassium. The data collected in diffused light condition work better than the data under direct sunlight.