A suboptimal midcourse guidance law is obtained for interception of free-fall targets in the three-dimensional space. Neural networks are used to obtain an on-board algorithm suitable for real-time implementation. Using the fact that optimal missile motion can be approximated by that on the vertical plane, the design of the guidance law in the three-dimensional space is implified with two commands, one using neural networks to generate optimal performance on the vertical plane and the other using feedback linearization to confine the missile motion within the plane. Also, only the motion on the vertical plane is considered in computation of time-to-go of the missile. Sensitivity to the neglected horizontal motion is investigated by omputer simulation.