The aim of this paper is to demonstrate the possibilities of estimating the track condition using axle-boxes and car-bodies motions described by acceleration signals. In the paper, the results presented indicate the condition of tracks obtained from the preliminary investigation on the test track. Furthermore, the results from the supervised runs (on Polish Railway Lines) of Electric Multiple Unit (EMU-ED74) with the prototype of track quality monitoring system installed on-board are described. As Track Quality Indicator (TQI) algorithm, used in the mentioned prototype, a modified Karhunen–Loève transformation is used in preliminary preparation of acceleration signals. The transformation is used to extract the principal dynamics from measurement data. Obtained results are compared to other methods of evaluating the geometrical track quality, namely methods, which apply the synthetic coefficient Jsynth and five parameters of defectiveness W5. The results from the investigation showed that track condition estimation is possible with acceptable accuracy for in-service use and for defining cost-effective maintenance strategies.