Abstract: Under the action of glucose oxidase and peroxidase, the glucose in tobacco sample solution could make the quantitative condensation between chromogenic oxygen receptor 4-aminoantipyrine and phenol to generate red quinonoids. The color of the solution was transformed into gray value signal by Image software, so as to determine glucose in the sample. The sample was dried and ground, and an aliquot (0.1±0.000 1)g was added into 20 mL of water, then extracted by vibration at (30±0.5)℃ for 30 min. After filtration with rapid qualitative filter paper, the initial filtrate was discard, and 2-3 mL of the subsequent filtrate was collected for use. The above filtrate (100 μL) was transferred into a 2 mL-microcentrifuge tube equipped with R1 and R2 reagents of 450 μL each, mixed well, and the mixture was placed into a constant temperature water bath shaker at (37±0.5)℃ for 5 min. Smart phone was used to take photo of the color solution, then the color was transformed into gray mode by Image software to obtain gray value. Using water as reference, the gray value of the color solution was normalized to obtain the gray value ratio. The standard curve was drawn with the logarithm of concentration of glucose as horizontal coordinate and the gray value ratio as vertical coordinate. It was shown that the linear range of the standard curve was 0.100-4.000 mmol·L^-1, and the detection limit (3s) was 0.021 6 mmol·L^-1. The 9 kinds of carbohydrate standard solutions, including glucose, fructose, mannose, xylose and so on with same concentration were determined. Except for glucose, the gray values of the other carbohydrate color solutions were not significantly different from the blank solution. Tests for precision and recovery on the standard flue-cured tobacco samples showed that RSDs (n=6) of the determined values were not more than 4.0%, and the spiked recoveries of glucose were in the range of 98.3%-103%. This method was applied to the analysis of actual samples, and compared with high performance liquid chromatography. The results showed that there was no significant difference between the two methods at 95% confidence.