An automatic assembly planning program should be able to analyze the assembly kinematic constraints, generate possible motions and select the appropriate assembly machines. This paper develops an approach for evaluating a robot for assembly tasks and planning its motions. The kinematical characteristics of both the robot and the assembly tasks are described by infinitesimal screws. The possible motions of the assembled parts are represented by convex cone of screws derived from the geometrical constraints. The robot’s possible motions are represented by the Jacobian cone, which is a modification of the Jacobian matrix that provides a common mathematical basis for convex cones operations. The selection procedure of the robot is based on the intersection of its cone with the assembly cones which results in feasible assembly directions for the robot. Choosing the optimal assembly direction and the robot posture is obtained by minimizing a cost function subjected to the constraints of both the robot and the assembly. Several given examples demonstrate the evaluation procedure and the planning approach.