A robotic arm able to reproduce the tremor characteristics of Parkinson Disease (PD) and Essential Tremor (ET), both in time and frequency domain, is presented. To obtain the tremor signals, an acquisition system with inertial sensors has been developed and implemented in affected patients. Different strategies for digital signal processing were considered to characterize the tremor. Neural networks demonstrated to be the best fit for modeling the highly non-lineal behavior of the analyzed system. The constructed mechatronic simulator was thought as a dynamic tool for developing and proving new approaches for active cancelation of tremor (ACT) as a hope for improving everyday life of people suffering from this disability.