The paper considers the development and implementation of a near-time-optimal neural-network-based position control method for DC motor servosystems. To bypass the difficulties caused by system constraints and modelling uncertainties, the paper uses classification neural networks to learn the time-optimal control law from experimentally generated near-time-optimal trajectories. In addition, by using regression neural networks to learn the relationship between control object displacement and the armature voltage pulse-width, a variable-pulsewidth control strategy is developed to achieve accurate positioning. Experimental results are given to demonstrate the effectiveness of the proposed approach.