In this paper we present a micro-electrical mechanical system (MEMS) based pedestrian navigation system (PNS) for seamless positioning. The sub-algorithms for the PNS are developed and the positioning performance is enhanced using the modified receding horizon Kalman finite impulse response filter (MRHKF). The PNS consists of a biaxial accelerometer and a biaxial magnetic compass mounted on a shoe. The PNS detects a step using a novel technique during the stance phase and simultaneously calculates walking information. Step length is estimated using a neural network whose inputs are the walking information. The azimuth is calculated using the magnetic compass, the walking information and the tilt compensation algorithm. Using the proposed sub-algorithms, seamless positioning can be accomplished. However, the magnetic compass based azimuth may have an error that varies according to the surrounding magnetic field. In this paper, the varying error is compensated using the MRHKF filter. Finally, the performance enhanced seamless positioning is achieved, and the performance is verified by experiment.