Current and future directions in the development of numerical methods and numerical software for control problems are discussed. Major challenges include the demand for higher accuracy, robustness of the method with respect to uncertainties in the data or the model, and the need for methods to solve large scale problems. To address these demands it is essential to preserve any underlying physical structure of the problem. At the same time, to obtain the required accuracy it is necessary to avoid all inversions or unnecessary matrix products. We will demonstrate how these demands can be met to a great extent for some important tasks in control, the linear-quadratic optimal control problem for first and second order systems as well as stability radius and H-infinity norm computations. (C) 2003 Elsevier Science B.V All rights reserved.