The study aimed to clarify the VIS/NIR spectral characteristics of Cabernet Sauvignon leaves with phosphorus deficiency, then to construct a rapid and nondestructive diagnosis model, which is expected to help the vineyard management and disease control. Firstly, the grape leaves in healthy, early and later stress by phosphate deficiency were analyzed by VIS/NIR micro fiber spectrometer. In order to remove noise interference, four preprocessing methods, including Savitzky-Golay convolution smoothing (S-G Smoothing), moving average smoothing (MAS), standard normal variate (SNV) and multiple scattering corrections (MSC), were used to optimize spectral signals. Then, the successive projections algorithm (SPA) was used to select the feature wavebands of leaf phosphate deficiency. Finally, the support vector machine models were constructed based on four different kernel functions, including linear kernel function (Linear), polynomial kernel function (Poly), radial basis function (RBF) and Sigmoid tanh function (Sigmoid), to diagnose the phosphate deficiency of leaves. The sensitivity (SEN) and accuracy (CCR) were cited to assess the availability and effectiveness of those models. Experimental results proved that S-G Smoothing was the best preprocessing method because of the better signal-to-noise ratio of spectrum processed by it and the good availability of the model based on it. Principal component analysis (PCA) was used to find outliers with a confidence interval of 95%. 22 samples were identified with outliers and removed. Eleven wavebands (402.6, 404.6, 409.0, 411.5, 539.4, 691.9, 729.9, 838.7, 1 011.9, 1 017.5 and 1 020.5 nm) were selected by SPA to consider as reflecting the information of phosphate deficiency and be the input variables of the diagnosis model. After the contrast of four models with different kernel functions, it can be known that the SVM model with Linear showed better sensitivity and accuracy than others. Its SEN was 81.08%, and CCR was 100% for healthy leaves, its SEN was 100%, and CCR was 84.78% for early-stage diseased leaves, and its SEN and CCR were 100% for late-stage diseased leaves. In this study, A rapid and nondestructive diagnosis method was proposed based on VIS/NIR spectroscopy for phosphate deficiency of the Cabernet Sauvignon leaves, which is expected to improve the management and disease control of the vineyard and the intelligence of wine grape cultivation.