The direct neural network control is introduced and the dynamic feedback recurrent neural network (RNN) with multi-layers as a controller is analyzed. A new type of RNN control system is proposed with applications to a continuous stirred tank reactor (CSTR) control and a plant wide industrial process control (Benchmark) problems. The control results show that the RNN controller with multi layers leads to simplification of control structure and good effect, especially for multi input/multi-output cascade control systems, exhibiting that the RNN controller is adaptable and applicable to complex industrial problems.