This paper discusses the 3D affine motion estimation problem using two cameras via observations of a single feature point. The unknown parameters to be estimated include the nine rotational parameters, the three translational parameters, and the 3D position. One camera assumes a parabolic projection. The other camera is the conventional camera that has a planar projection surface. The parabolic camera is needed due to its ability of estimating the nine rotational parameters independently of the other camera. The conventional camera is utilized to solve the depth ambiguity problem. A closedloop nonlinear observer is developed for the affine motion estimation problem. Simulation results are presented to show the robustness of the proposed motion estimation scheme. Key Words: Motion estimation, affine motion, parabolic projection, stereo vision.