This paper presents an evolutionary algorithm for collision-free path planning of multi-arm robots. A global path planning technique is used to plan paths for two robots. The paths are represented by a string of via-points which the robots have to pass through. The path planning algorithm uses evolutionary techniques to minimise path lengths, uneven distribution of via-points on the paths to minimise variation in velocity and to minimise the collision between the robots. This algorithm is different from previously reported GA (genetic algorithms) based path planners in that the via-points are not limited to lie on points on a fixed grid over the workspace of the robot and have not been encoded as binary numbers. Simulation results are presented for collision-free paths planned for two planner arms and then for two 3-DOF PUMA-like arms moving in three dimensional operational space.