Abstract: Identification modeling and quantitative assessment of minimum step size dynamic characteristics in CNC machine tools were inveatigated to resolve difficulties in valid data identification and the lack of objective evaluation metrics. Initially, an effective data identification model was developed by leveraging the time-series characteristics and weight distribution of the test data, and global optimization was achieved through the integration of feature engineering and a grid search algorithm, thereby enabling the efficient and precise extraction of valid data. Building upon this foundation, a comprehensive evaluation framework was established across three dimensions, namely curve consistency, positioning error, and reversal error. Curve consistency was quantified by the root mean square error ratio (RMSE-ratio), which was employed to assess the goodness of fit between the measured trajectory and the ideal step response, thus evaluating the machine tool's capability to execute the smallest programmed command. The positioning error was characterized by the maximum single-step error and the maximum cumulative error, both of which were extracted to represent the displacement accuracy of the actuator. Reversal error was quantified by computing the area enclosed by the hysteresis loop formed from the ascending and descending step-length response curves, thereby capturing the system’s reversal behavior and hysteresis performance. Collectively, the dynamic response characteristics of the machine tool under minimal step movements were intuitively and accurately quantified, and an objective, reproducible basis was provided for the comparative evaluation of performance across various motion components.