Measuring or inferring the physiological properties of motor neurons, such as during disease progression or aging, remains challenging, often requiring longitudinal invasive measurements or analysis techniques based on simplifying assumptions. Here we use the framework of simulation-based inference to train neural density estimators that directly infer the posterior distribution of properties of interest (i.e., the physiological properties most likely to explain the observations) by simulating from a state-of-the-art electromyography simulator. We not only surpass conventional methods in accuracy and sensitivity, but also infer properties that have so far been impossible to measure. We believe this will significantly impact the possibilities for both clinical and research contexts in motor neurophysiology.