Track maintenance and preventive maintenance are necessary for the safe and comfortable running of railways. However, there is a problem that sufficient inspections cannot be performed by local railway operators with limited personnel and financial resources. In response to this problem, a track condition monitoring system was developed and applied to regional railways in Japan. In the developed system, car-body vibration is used to analyse the track condition. However, track maintenance is based on measured track irregularity at millimetre. As the car-body acceleration is not clearly linked to the track irregularity, it is difficult to use the car-body vibration for track management. Thus, it is necessary to develop a method to estimate track irregularity from car-body vibration. In this paper, we proposed a new technique for estimating track irregularities from measured car-body acceleration. Gaussian process regression was applied to the simulated data, and the obtained response surface was used to estimate track irregularities. An application example for a regional railway line showed that the proposed method is useful for track irregularity management.