An identification scheme to adaptively identify the time-varying amplitude of principal modal components from the time domain operating deflection shapes (ODS) is studied. The ODS depend on their excitation sources and have fixed quantitative values. Since the mode shapes cannot be used to synthesize the ODS, the actual forcing vector applied to the system usually cannot be identified. Therefore, measurement of the time domain ODS requires a set of vibration responses obtained from a multichannel data acquisition system under actual operating conditions. However, the responses are highly affected by noise. First, we present an identification scheme of the time-varying amplitude of principal modal components based on both principal component analysis and the total least squares method. Then we adapt the scheme for application to time-varying systems, making use of URV decomposition. The resulting scheme can adaptively decompose the deformation of a structure into the underlying superposition of several modes. We demonstrate its performance on artificially generated response data. The results show that this scheme can be used to accurately provide estimate the time-varying amplitude.