The invention discloses a multispectral remote sensing data supervised visualization method. The method includes 1. selecting each land type sample area on a grid image and extracting the gray value of each wave band of the corresponding pixel to form a sample; 2. calculating the main component transformation coefficient by means of the sample grey values An (a0, a1, ...a11); 3. performing the regression analysis and calculation on the first, second and third main components and the land type main component mean values to obtain a new transformation coefficient Cn(c0, c1,...c11); and 4. applying the coefficient Cn(c0, c1, ...c11) to the original grid data to obtain the new visualized image. The present invention relates to a method for realizing visualization by means of learning samples to carry out the supervised characteristic generation, intra-class variance reduction and noise effect elimination on the basis of the principal component transformation. It is an object of the present invention to explore a method for generating visualization features based on learning samples such that visualization is combined with visual interpretation applications. The method can reduce the difficulty of visual interpretation and provide a new method for visualizing remote sensing data.