A quantum neural networks model and its algorithm are proposed. The model is a group of the quantum gate
circuits. The input information is expressed as the qubits, which, as the control qubits after rotated by the rotation gate,
control the qubits in hidden layer to reversal. The qubits in hidden layer, as the control qubits after rotated by the rotation
gate, control the qubits in output layer to reversal. The final output is described by the probability amplitudes of excited states
in output layer. The learning algorithm is presented based on the gradient descent algorithm. The simulation results show
that the proposed algorithm is superior to the common BP neural networks in both convergence capability and robustness.