An investigation is presented into the neural networks based active vibration control of flexible linkage mechanisms. A smart mechanism featuring piezoceramic actuators and strain gauge sensors is designed. A nonlinear adaptive control strategy named Neural Networks based Direct Self-Tuning Control (NNBDSC) is employed to suppress the elastodynamic responses of the smart mechanism. To improve the initial robustness of the NNBDSC, the Dynamic Recurrent Neural Network (DRNN) controllers are designed off-line to approximate the inverse dynamics of the smart mechanism. Through on-line control, the strain crest of the flexible link is reduced 60 percent or so and the dynamic performance of the smart mechanism is improved significantly.