Abstract.Goodman and Saff conjectured that if / is convex in the direction of the imaginary axis then so are the functions \f(rz) for all 0 < r < \2 -1, i.e., the level sets f(\z\ < r) are convex in the direction of the imaginary axis for 0 < r < \¡2 -1. A weak form of this conjecture is proved and a question of Brannan is answered negatively.Let Ur= [z:\z\ < r) and let 5 denote the class of all functions f(z) = z + a2z2 + ■ ■ ■ analytic and univalent in U = Uv For / e S there are several geometric properties possessed by f(U) that are inherited by its level sets Gr = f(Ur) for all 0 < r < 1. For example if f(U) is either convex, starlike, or close-to-con vex, then so are its level sets Gr for all 0 < r < 1.An analytic function / is said to be convex in the direction of a line Le: te'8 (-co < t < oo) if the intersection of f(U) with each line parallel to Le is either a connected set or empty. Let CIA denote those functions / for which f(U) is convex in the direction of the imaginary axis with /(0) = 0 and /'(0) = 1. Since CIA functions are close-to-convex, they are univalent. It came as a bit of a surprise when Hengartner and Schober [4] constructed an example where / e CIA but the corresponding level sets Gr were not convex in the direction of the imaginary axis for all r sufficiently close to 1. A more quantitative result was obtained by Goodman and Saff [3]. They were able to prove that for each ^2 -1 < r < 1 there exists an /e CIA for which |/(rz) £ CIA. Hence they conjectured that if /e CIA then jf(rz) e CIA for all 0 < r < y2 -1, i.e., the level sets Gr are convex in the direction of the imaginary axis for 0 < r < y2 -1.In [1, Problem 6.53] Brannan asked whether or not: If / e CIA and 7-J(rQz) £ CIA for some 0 < r0 < 1, does this imply that jf(rz) £ CIA for all r0 < r < 1? This question was motivated by the example constructed by Hengartner and Schober [4]. In this note we prove a weaker form of the Goodman-Saff conjecture and answer Brannan's question. Our main result is the following theorem.Theorem. /// e CIA then there exists a set I c [0, tr] of positive measure such that jf(rz) is convex in the directions L8: te'6 (-oo < t < oo) for all 0 < r < y2 -1 and all 0 e /.