Attitude determination systems utilizing low cost MEMS sensors are increasingly becoming important due to its advantages in terms of the quickly improved precision, robust, high dynamic response and more significantly inexpensive costs of development and usage. However the large noises inherent in low cost MEMS sensors degrade the derived attitude precision if utilized through the conventional methods, e.g. initial alignment, strapdown inertial navigation mechanization. Therefore the novel application approach suitable for MEMS needs to be investigated. This paper describes an attitude determination system that is based on low cost MEMS inertial sensor, a triad of magnetometers and a commercial GPS receiver. Two main issues are addressed in the paper; firstly determination of the attitude initials, the algorithm is based on a quaternion formulation, a representative of attitude, of Wahbapsilas problem, whereby the error quaternion becomes the estimated state and is corrected by two observations of the earth magnetic field and gravity respectively. After the estimates converge, the derived attitude parameters are employed to initialize the inertial navigation calculations. Due to the large noises in MEMS sensor, there is a demand for external velocity and/or position corrections in the MEMS navigation calculations when system experiences translational motions. Hence secondly, GPS solutions are integrated in a Kalman filter by providing external velocity and position observations. A Kalman dynamic model is designed appropriate for MEMS sensor noise characteristics. The bias and drift are estimated by the integrated Kalman filter, which enables the online calibrations of MEMS sensor. The proposed approach has been developed and its efficiency is demonstrated by various experimental scenarios with real MEMS data and they are compared with Novatel SPAN-IMU reference.