The surface condition of bike lanes is one of the main factors that cyclists consider when choosing their route, because it affects their travelling comfort, inasmuch as it is related to the vibrations they experience while riding. The existing alternatives to determine this condition either do not correspond to the cyclist's reality or their application is complex. A methodology based on a logistic regression is proposed, which is objectively and specifically aimed at detecting bike-lane pavement failures and thus inferring its condition in the analyzed section. As a proof of concept of the proposed methodology, a field experiment was designed, which emulates two specifics defects: pavement upheavals and potholes. Data collection was carried out by attaching inertial sensors to the test bicycle. The application of the proposed methodology allowed identifying the necessary inertial variables to assess the considered failures. Among the latter, the main ones are rotations in the three axes and vertical acceleration. Four models that could correctly identify pavement issues were generated with these variables. In the future, the results thereof will allow building an indicator to infer the surface condition based on the vibrations felt by cyclists, and thus establish a level of service associated to the bike-lane pavement through these indicators.