This research investigates a novel data-driven approach to condition monitoring of electromechanical actuators (EMAs) consisting of feature extraction and fault classification. The approach is able to accommodate time-varying loads and speeds since EMAs typically operate under nonsteady conditions. The feature extraction process exposes fault frequencies in signal data that are synchronous with motor position through a series of signal processing techniques. A resulting reduced dimension feature is then used to determine the condition with a trained Bayesian classifier. The approach is based on signal analysis in the frequency domain of inherent EMA signals and accelerometers. For this work, two common failure modes, bearing and ball screw faults, are seeded on a MOOG MaxForce EMA. The EMA is then loaded using active and passive load cells with measurements collected via a dSPACE data acquisition and control system. Typical position commands and loads are utilized to simulate “real-world” inputs and disturbances and laboratory results show that actuator condition can be determined over a range of inputs. Although the process is developed for EMAs, it can be used generically on other rotating machine applications as a Health and Usage Management System (HUMS) tool.