It is necessary to reduce steady increases in wheel load fluctuation and rolling noise with increases of the maximum speed of Shinkansen. The cause is mainly considered in short wave track irregularities, such as rail surface roughness or a loose sleepers. However, the Track Geometry Car, which is now working for daily inspection to check longitudinal levels, can not detect such short wave track irregularities. So, we take notice of axle box acceleration to detect these defects. Data analyses are carried out to determine the relationship between axle box acceleration and short wave track irregularities, and to make clear the statistical characteristics of axle box acceleration. Finally the cause of defects is identified through several filter banks. This paper describes the results of these analyses, and application of new analysis methods, such as wavelet analysis to axle box acceleration.