Abstract: As a solution to the large fluctuation of moisture content in output tobacco and time delay associated with the traditional PID control mode of loosening and conditioning procedure, a principal component neural network prediction model for moisture content in output tobacco was proposed based on BP neural network and principal component regression method. To minimize the prediction error, a genetic algorithm-based optimization model for water addition during loosening and conditioning and a real-time feedback mechanism for correcting the water addition were established to obtain the optimal water addition with self-adaptive model parameters. The prediction model was tested on blended tobacco material for "Furongwang (Hard)" brand cigarette in Lingling Cigarette Factory of China Tobacco Hunan Industrial Limited Corporation. The results showed that: 1) The average absolute error between the predicted and actual moisture contents was 0.14%, the variation trend of moisture content in output tobacco could be well predicted. 2) Compared with the traditional PID control mode, the principal component neural network control model increased the mean process capability index of moisture content in output tobacco by 65% and decreased the mean standard deviation by 37%. This technology is helpful for promotion of control accuracy and stability of moisture content in output tobacco from the loosening and conditioning procedure.