Conventional modal testing is known as a powerful tool for dynamic analysis of structures. However, for some engineering structures, conventional modal testing is difficult or even impossible to conduct due to the problems associated with the artificial excitation of structure. Operational Modal Analysis (OMA) is one solution to deal with these cases. In OMA the structure is excited by ambient forces and only the responses are measured and taken into account. Accelerometers are the traditional tools for measuring the responses of structure. It is well khonwn that the measured responses are contaminated by bias errors corresponding to the mass-loading effect of accelerometers. This causes the natural frequencies of structure are measured lower than the real values. In this paper a new method is proposed for eliminating the mass-loading effects of accelerometers from measured responses in OMA. A numerical model of a mass-spring-damper system is used for validation of the method. Also a steel plate is used for experimental validation of the proposed approach. The results are confirmed by those of conventional modal testing. Both numerical and experimental results show that the proposed method can effectively eliminate the mass-loading effects of accelerometers from measured responses in OMA. Also, the method has the ability to correct the measured natural frequencies from OMA accurately.