We examine discrepancies in the angular momentum budget of the Earth on subannual timescales. We find that for signals with long periods (>30 days) the phase of the length of day (ΔLOD) significantly leads that of the atmospheric angular momentum (AAM). The discrepancy has an amplitude of ≃15% of the observed AAM at periods of around 100 days. We show that if changes in the ΔLOD are driven by the atmosphere, then this is diagnostic either of the importance of the oceans or of dynamical coupling between the core and mantle for these frequencies. We find that a simple model of core-mantle coupling can cause a phase lead of ΔLOD of the correct magnitude. We attempt to fit the transfer function with a simple three-layer model of the Earth, treating the core as a rotating solid body coupled to the mantle by a specified function. Although our simple model can supply sufficient phase lead of the ΔLOD, the fit to the frequency dependence is poor and it can not fit the magnitude of the transfer function. Nevertheless, this suggests that core-mantle coupling is a plausible explanation of the observed phase difference. Better fits might be obtained with more realistic models of the core.¿ 1997 American Geophysical Union