Fall is a risky event in the elderly people’s daily life, it often cause serious injury both in physiology and psychology. A MEMS attitude measurement system is designed for fall detection in real time. This paper presents the design and error test of the attitude measurement unit. Each unit contains orthogonally mounted triads of accelerometers, magnetometers and gyros. With an integrated microcontroller for attitude calculating and flash for data storage, the size of the unit is 32mm×23mm×12mm. An extended Kalman filter based on quaternions is designed for attitude measurement. The digital angle output rate is 100Hz. A new method based on coordinate transformation for attitude measurement error test is introduced, using a single axis turntable and a fixed angle wedge. Theory of the testing method is presented and test experiments are performed. Test results show that attitude measurement error is less than 2°, which meets the requirement of fall detection precision. The fall detection system consists of five attitude measurement units fixed on the human legs and waist.