A new method was proposed for early diagnosis of disease and pest level based on stem water content, which provided early warning for diseases and pests. Lagerstroemia indica seedlings with different health levels were monitored for acquiring stem water content. Then the features of stem water content were respectively extracted by two methods, including key parameter and principle component analysis. Ultimately, some supervised and unsupervised models were established for early diagnosis of disease and pest level on Lagerstroemia indica. Judging from variance analysis, the effects of health level on four key parameters (daily minimum, maximum, average and range of stem water contents) were all in very significant difference. Judging from principle component analysis, the cumulative contribution rate of the first four principal components of stem water content reached 99.7%. Among supervised models, BP model with input of PCA features performed the best and its average recognition reached 98%. Among unsupervised models, Kmeans model with input of PCA features performed the best and its average recognition rate reached 92%. Hence, stem water content can be chosen as a reliable index for early diagnosis of plant disease and pest level. The PCA features were superior to the key parameter features. The performance of supervised models was better than that of unsupervised models.