The work aims to improve the micro-liquid filling accuracy and design a self-tuning control system with re-filling process as research object. The basic structure, technological process as well as automatic re-filling principle of filling machine were introduced. A secondary re-filling controller based on fuzzy RBF neural network was designed. The neural network structure and learning algorithm were also expounded. A corresponding control system was constructed based on PLC and ARM. The PLC was main controller, and it was responsible for sensor signal detection and working procedure implementation. The ARM was subordinate controller, and it was responsible for secondary re-filling control. Finally, the experimental research was done. Comparison results showed that the filling accuracy was improved obviously and the error ratio could be controlled below 1%. The described control system can minimize the production process error and meet the technological requirements of the filling.