【Objective】To explore relationship between color of sandalwood leaves and content of total iron in the plant, a visible-light-spectrum-based sandalwood image segmentation method was bought forth for prediction of content of total iron in the plant.【Method】 First of all, Otsu’s method was used to remove the pigments of soil and the other green plants, by segmenting Channel b, and then Channel L was extracted, and again Otsu’s method was used to extract the image of sandalwood out of its background. Then burrs of the image were smoothened through median filtering and morphological operation. Based on the fact that new and old leaves varied differently in color under iron stress, a method for determination of new and old leaf ratio was developed. First, the minimum circumcircle of the segmented sandalwood was to be defined, and then calculation was done of the ratio of the canopy breadth measured last time to that measured this time, and then the ratio was multiplied by the radius of the minimum circumcircle to gain radius of the concentric circle. The ring part between the two concentric circles represented new leaves and the rest old leaves. Color value of each channel (R, G, B, H, S, I, L, a and b) was calculated. Then four groups of comparison were designed (spectral value of the whole plant, spectral value of new leaves, ratio of the spectral values of new leaves and the whole plant, and ratio of spectral values of new leaves and old leaves). And in the end, predictions of the content of total iron were analyzed using the BP neural network modified with different methods.【Results】(1) The segmentation algorithm proposed in this paper is better than the support vector machine in result, with pixel error ranging within 5%, and the errors of all RGB channels controlled within 3%. (2) The optimum content of total iron in sandalwood leaves varies between 250~300 mg kg^-1. When the content of total iron in leaves is less than the optimum value, the color value of Channel G increases while that of Channels R and B decrease with rising content of total iron. But when the content of total iron in leaves gets beyond the optimum value, the trend goes reversely, which indicates that being either too high or too low iron content would be a factor causing chlorosis in leaves. (3) Comparison shows that the prediction based on the ratio of spectral values of new leaves and old leaves is the best, while that based on the spectral value of the whole plant, the worst, which indicates that the method, proposed in this study, of comparing new and old leaves in spectral value is the most effective one, reflecting the content of total iron in the plant. And (4) In terms of efficiency and effectiveness, the four kinds of neural network models exhibits an order of GA-BPNN > PSO-BPNN > BPNN-Adaboost > BPNN, which indicates that optimization is better than the iteration, and that appropriate initial value and threshold value have more influence on prediction ability of the neural network models .【Conclusion】All the findings of this research have a guiding significance for nutritional diagnosis of precious tree species in terms of micro-elements, and provide a new way of thinking for precision forestry.