Purpose <p>Modal identification of time-varying structures is a fundamental aspect in the field of structural dynamics, from which the identified modal parameters are considered as important indicators for damage identification and structural condition assessment in structural health monitoring. Therefor a nonparametric modal identification approach for time-varying structures from a data-driven perspective is investigated.</p> Methods <p>The inherent connection between dynamic mode decomposition and operational modal identification is revealed, where the modal identification is converted to solve the eigenvalues of dynamic matrix in dynamic mode decomposition. Furthermore, the full implementation procedures of the method for time-varying modal parameters extraction are provided, where the dynamic mode decomposition and sliding-window technique is effectively combined. Finally, a series of numerical examples are performed to validate the effectiveness and practicability of the method, where the dynamic modelling of simply supported beam with moving mass is derived and the classical methods based on signal decomposition for time-varying modal identification are used for comparison.</p> Results <p>The results from studied cases show that the proposed method has the capability for accurately identifying the instantaneous modal frequencies with good efficiency when the external excitation is unknown, showing promising prospects for engineering applications.</p> Conclusion <p>An online modal identification method for time-varying structures based on dynamic mode decomposition and sliding-window technique is presented without requiring establishing an accurate parameterized mathematical model. The proposed method can well identify instantaneous modal frequencies, and provides a more easily implementable framework with few adjustable parameters and lower computational resources and better identification performance compared with the VMD-HT and STMVMD method.</p>