This paper investigates discriminating tongue features to distinguish between early stage BC patients and normal persons via non-invaded methods, expecting to detect BC in the early stage and give treatment in time to increase the recovery rate and lower relapse rate. The tongue features for 67 breast cancer patients of 0 and 1 stages, and 70 normal persons are extracted by the Automatic Tongue Diagnosis System (ATDS) [4-6, 28-31]. A total of nine tongue features, namely, tongue color, tongue quality, tongue fissure, tongue fur, red dot, ecchymosis, tooth mark, saliva, and tongue shape are identified for each tongue. Features extracted are further sub-divided according to the areas located, i.e., spleen-stomach, liver-gall-left, liver-gall right, kidney, and heart-lung area. The purpose focuses on inducing significant tongue features (p<0.05) to discriminate early-stage breast cancer patients from normal persons. The Mann-Whitney test shows that the amount of tongue fur (p = 0.024), maximum covering area of tongue fur (p = 0.009), thin tongue fur (p = 0.009), the average area of red dot (p = 0.049), the maximum area of red dot (p = 0.009), red dot in the spleen stomach area (p = 0.000), and red dot in the heart-lung area (p = 0.000) demonstrate significant differences. Next, the data collected are further classified into two groups. The training group consists of 57 early-stage breast cancer patients and 60 normal persons, while the testing group is composed of 10 early stage breast cancer patients and 10 normal persons. The logistic regression by utilizing these 7 tongue features with significant differences in Mann-Whitney test as factors is performed. In order to reduce the number of tongue features employed in prediction, we remove one of the 7 tongue features with lesser significant difference (p>0.05) and perform logistic regression twice. In the first time, we remove the maximum area of red dot (p = 0.266), and perform logistic regression. Among them, the amount of tongue fur (p = 0.000), the maximum covering area of tongue fur (p = 0.000), thin tongue fur (p = 0.008), the average area of red dot (p = 0.056), red dot in the spleen-stomach area (p = 0.005), red dot in the heart-lung area (p = 0.011) reveal independently significant meaning. In the second round, the average area of red dot (p = 0.056) is removed. The logistic regression shows that the amount of tongue fur (p = 0.001), the maximum covering area of tongue fur (p = 0.000), thin tongue fur (p = 0.007), red dot in the spleen-stomach area (p = 0.006), red dot in the heart-lung area (p = 0.003) reveal independently significant meaning. The tongue features of the testing group are employed in the aforementioned three models to test the power of significant tongue features identified in predicting early-stage breast cancer. An accuracy of 80%, 80% and 90% is reached on normal peoples by applying the 7, 6 and 5 significant tongue features obtained through Mann-Whitney test, respectively, while 60%, 60% and 50% is reached on the corresponding early-stage breast cancer patients.