Pure milks, such as Yili，Xinxiwang，Meilijian，Mengniu and Guangming, were selected and detected by an electronic tongue. The results of one-way analysis of variance showed that brands had a very obvious effect on sensors signals of pure milks. Sensors array was optimized by removing sensors variable of small F and R2. Principal component analysis(PCA) was applied to original data and optimized data(without BA sensor; without BA,BB sensors; without BA,BB,HA sensors; without BA,BB,HA,GA sensors). Results indicated that the optimized data without BA,BB,HA,GA sensors was more effective than the other data in distinguishing the brands of pure milks. Stepwise discriminant analysis was used to recognize，the identification rates of all data in the calibration sets were 100%. In the prediction sets，the identification rates of optimized data without BA,BB,HA,GA sensors and without BA sensor were both 90%，but the optimized data without BA,BB,HA,GA sensors only included three sensors variables and showed the best identification effects on the brands of pure milks. One-way analysis of variance could optimize the sensors array of electronic tongue by removing unremarkable sensor signals and improve the recognition performance of electronic tongue. 