To solve the difficulty in establishing the mathematical model of the cigarette chemical composition and tobacco quality, an improved fuzzy clustering-based ensemble evaluation model for tobacco quality is proposed. The method first determined the differences in chemical components among tobacco samples, and to obtain consistency results between model classification and expert evaluation results, simulated annealing algorithm was used to optimize the existing fuzzy clustering algorithm, and base classifier was established. On this basis, multiple classification results for different sample sets by the classifiers were integrated using the AdaBoost, and the final tobacco quality evaluation models was formed. The contents of 7 kinds of chemical composition including total sugar, reducing sugar, total nitrogen, nicotine, potassium ion, chlorine ion and protein in 130 group of tobacco leaf were determined, contrast experiment is done by the improved fuzzy clustering method, neural network algorithm and fuzzy clustering algorithm, the results showed that the error detection rate of the improved fuzzy clustering method is 6.7%, indicating the improved method has the ability to recognize small sample data, and is superior to the other compared methods.