In this paper, the author proposes a new method to design SCARA robots for higher repeatability. First, the author outlines various procedures used in optimal robot design and then points out among the various performance indices those related to repeatability. The author adds some new criteria issued from the stochastic ellipsoid theory. Another innovative part of the paper is to take into account a task-oriented strategy during the design stage, meaning the possibility of adapting task orientation and location in the robot workspace. These concepts are applied to SCARA optimal design. The method described here consists of considering simultaneously robot geometry and joint repeatability, keeping both the reach and the total cost of the sensors constant. It results in an optimization problem with adimensional ratios, which then allows easy comparisons with existing SCARA. The results are surprising and give some clues to answer the underlying question: Are industrial SCARA designed for high repeatability?