Abstract. An algorithm is described for computing an interpolation spline of arbitrary but fixed degree which preserves the convexity of the given data set. Necessary and sufficient conditions for the solvability of the problem, some special cases and error estimations are given.1. Introduction. In some problems arising from science or engineering the solution of an interpolation problem with constraints is required. For example, if a convex data set is given then the interpolant should also be convex. Using standard techniques like polynomial or cubic spline interpolation the resulting interpolant is not convex in general (see [5]).In recent years convex interpolation by splines has been investigated in several publications (see [3]-[17]). Various possibilities were proposed to assure the convexity of an interpolation spline. They reach from additional conditions on the data set to additional knots of the interpolation spline.In this paper we consider convex interpolation by splines of arbitrary degree k > 3 with smoothness q, where 1 < q < [(k -l)/2]. As is known, the solvability of such an interpolation problem is equivalent to that of a special system of linear inequalities. Necessary and sufficient conditions are derived such that this system has a solution, and an algorithm is described to find all solutions of the system of inequalities. Furthermore, a sufficient condition is given which always assures the solvability of the system and which is easy to test. In this case, a solution can be found even with a reduced algorithm. The sufficient condition just mentioned is interpreted in different ways. Specifically, for a strictly convex data set it is possible to give a lower bound for the degree k such that the interpolation problem is solvable. For the developed method no additional spline knots are needed. In the last section, the important question of error estimation is investigated. Under certain assumptions, error bounds are derived for continuous and continuously differentiable underlying functions. The assumptions are related to the solvability of the convex interpolation problem, and in one case to the ratio of contiguous step sizes of the grid. In the differentiable case, the order of approximation is better than in the continuous one.