In order to deal with the problem that the system dynamics of mechatronic servo systems are difficult to be described with precise models, and the measurement of state information is also effected by noise, a nonlinear backstepping control approach based on command filters and neural networks is proposed. The approach can effectively compensate the influence of unmodeled dynamics and external disturbances on the mechatronic servo system. In this approach, a command filter is introduced to acquire the differential estimation of known signals and cope with the noise. And then a neural network is applied to approximate the unknown system dynamics, including the unmodeled friction and the external disturbance. The update law of neural network weights is implemented online by using the gradient descend algorithm, without an off-line learning phase. Finally, the closed-loop system stability is rigorously proven by using the Lyapunov-based method. To verify the effectiveness of the proposed algorithm, extensive comparative experiments are implemented in the mechatronic servo experimental platform. The experimental results show that the proposed controller achieves satisfactory performance and the closed-loop system obtains satisfactory robustness with respect to system uncertainties and external disturbances.