In the medical field, there is a need for small ambulatory sensor systems for measuring the kinematics of body segments. Current methods for ambulatory measurement of body orientation have limited accuracy when the body moves. The aim of the paper was to develop and validate a method for accurate measurement of the orientation of human body segments using an inertial measurement unit (IMU). An IMU containing three single-axis accelerometers and three single-axis micromachined gyroscopes was assembled in a rectangular box, sized 20 x 20 x 30 mm. The presented orientation estimation algorithm continuously corrected orientation estimates obtained by mathematical integration of the 3D angular velocity measured using the gyroscopes. The correction was performed using an inclination estimate continuously obtained using the signal of the 3D accelerometer. This reduces the integration drift that originates from errors in the angular velocity signal. In addition, the gyroscope offset was continuously recalibrated. The method was realised using a Kalman filter that took into account the spectra of the signals involved as well as a fluctuating gyroscope offset. The method was tested for movements of the pelvis, trunk and forearm. Although the problem of integration drift around the global vertical continuously increased in the order of 0.50 degrees s(-1), the inclination estimate was accurate within 3 degrees RMS. It was shown that the gyroscope offset could be estimated continuously during a trial. Using an initial offset error of 1 rad s(-1), after 2 min the off-set error was roughly 5% of the original offset error. Using the Kalman filter described, an accurate and robust system for ambulatory motion recording can be realised.