We present a heuristic-based real-time reach planning algorithm for virtual human figures. Given the start and goal positions in a 3D workspace, our problem is to compute a collision-free path that specifies all the configurations for a human arm to move from the start to the goal. Our algorithm consists of three modules: spatial search, inverse kinematics, and collision detection. For the search module, instead of searching in joint configuration space like most existing motion planning methods do, we run a direct search in the workspace, guided by a heuristic distance-to-goal evaluation function. The inverse kinematics module attempts to select natural posture configurations for the arm along the path found in the workspace. During the search, candidate configurations will be checked for collisions taking advantage of the graphics hardware – depth buffer. The algorithm is fast and easy to implement. It allows real-time planning not only in static, structured environments, but also in dynamic, unstructured environments. No preprocessing and prior knowledge about the environment is required. Several examples are shown illustrating the competence of the planner at generating motion plans for a typical human arm model with seven degrees of freedom.