In the present paper we proceed the data-driven modeling of a two degrees of freedom (2-DOF) piezoelectric micromanipulator through models with the Nonlinear Au-toRegressive with eXogenous inputs (NARX) structure and real acquired data. We show the results when the system is excited at high frequencies, aiming towards rapid and precise micropositioning. The order of NARX the models are increased until they satisfy the statistical tests based on higher-order correlations and the multiple correlation coefficients, which are close to unity for both measured outputs. The results herein presented encourage the use of data-driven methods for modeling piezoelectric micromanipulators.