The simplicity of Transpose Jacobian (TJ) control is a significant characteristic of this algorithm for controlling robotic manipulators. Nevertheless, a poor performance may result in tracking of fast trajectories, since it is not dynamics-based. Use of high gains can deteriorate performance seriously in the presence of feedback measurement noise. Another drawback is that there is no prescribed method of selecting its control gains. In this paper, based on feedback linearization approach a Modified TJ (MTJ) algorithm is presented which employs stored data of the control command in the previous time step, as a learning tool to yield improved performance. The gains of this new algorithm can be selected systematically, and do not need to be large, hence the noise rejection characteristics of the algorithm are improved. Based on Lyapunov's theorems, it is shown that both the standard and the MTJ algorithms are asymptotically stable. Analysis of the required computational effort reveals the efficiency of the proposed MTJ law compared to the Model-based algorithms. Simulation results are presented which compare tracking performance of the MTJ algorithm to that of the TJ and Model-Based algorithms in various tasks. Results of these simulations show that performance of the new MTJ algorithm is comparable to that of Computed Torque algorithms, without requiring a priori knowledge of plant dynamics, and with reduced computational burden. Therefore, the proposed algorithm is well suited to most industrial applications where simple efficient algorithms are more appropriate than complicated theoretical ones with massive computational burden. (c) 2007 Elsevier Ltd. All rights reserved.