Force optimization of multifingered robotic hands can be formulated as a convex programming or non-convex programming problem. This paper presents a neural dynamical method for real-time optimal control of force distribution. Compared with existing approaches to the force optimization, the proposed neural dynamical approach is shown to have a lower model complexity and an exponential convergence even in the case of singular positions. A simulation example shows that the proposed neural dynamical approach can achieve optimal force distribution in real time.