Abstract.In this paper we discuss the recurrent task of evaluating a linear functional defined by (generally infinitely many) linear constraints. We develop a theory for the stability of this problem and suggest a regularization procedure, based on orthogonal expansions. Simple and efficient computational schemes for evaluating the functional numerically are given. As a particular instance of the problem (1) and (2) is uniquely determined by the sequence cr = L{ar), r = 1, 2, . . . .Proof. Let bn be the polynomial of degree less than n which approximates b best in the maximum norm. bn is uniquely determined and II b -bn II -► 0 when n -► °°. Hence L{b) = limn_>00Z-(ôn), and the conclusion follows. Q.E.D.However, by practical calculations cr are known only with a finite accuracy and only finitely,many of the conditions (2) may be taken into account. Hence, a certain error is introduced in the calculated value of L{b) which is determined by approximating b (directly or indirectly) with linear combinations of a,, a2, . . . , an. The purpose of this paper is to extend and generalize the results in [6] and [7] as well as to de-