Ambulatory assessment of trunk posture is important in improving our understanding of the risk of low back injury. Recently, small inertial sensors combining accelerometers, gyroscopes and magnetometers were developed and appear to be promising for measuring human movement. However, the validity of such sensors for assessing three-dimensional (3D) trunk posture in motion has not been documented. The purpose of this study was to evaluate a hybrid system (HS) composed of two inertial sensors for the 3D measurement of trunk posture. A secondary purpose was to explore the utility of adding another source of information, a potentiometer, to measure the relative rotation between both sensors in order to improve the validity of the system. The first sensor was placed over the sacrum and the second on the upper part of the thorax. Both sensors were linked by a flexible rod with a potentiometer. A complementary quaternion filter algorithm was used to estimate trunk orientation by taking advantage of the nine components of each sensor and the potentiometer. The HS's orientations were compared to those obtained from a 3D optoelectronic system. Validation of the HS was performed in three steps in which six subjects had to perform manual handling tasks in: (1) static postures; (2) dynamic motions of short duration (30s); and (3) dynamic motions of long duration (30min). The results showed that the root mean square (RMS) error of the HS was generally below 3 degrees for the flexion and lateral bending axes, and less than 6 degrees for the torsion axis, and that this error was lower for the short-duration tests compared to the long-duration one. The potentiometer proved to be an essential addition, particularly when the magnetometer signals were corrupted and only the gyroscope and accelerometer could be combined. It is concluded that the HS can be a useful tool for quantifying 3D trunk posture in motion.