We have analyzed the long-period (~1000 s) part of the Earth's free oscillation spectrum of the 1989 Macquarie Ridge earthquake. This event stands out as the largest in the last decade and as the first great earthquake to have been recorded digitally on the new generation of very broadband seismic sensors around the globe. Its spectrum was modeled for five spectral peaks, 0S4, 0S6, 0S9, 1S8, and 5S3, taking into account the effects of lateral heterogeneity, rotation, and ellipticity which cause coupling within the spectral peaks (self coupling) and between different multiplets (multiplet coupling). The effects of self coupling were calculated following the formulation of Woodhouse and Girnius (1982). Multiplet coupling only weakly affected the spectral peaks that were used. A large discrepancy in phase and amplitude between the calculated spectrum and the data was observed, which suggests that the seismic moment was ~50% larger than that inferred from surface wave analysis (Satake and Kanamori, 1990). An error analysis was performed in order to detect inaccuracies that might have resulte in this discrepancy; centroid time, source finiteness correction, and directivity effects are likely to cause a significant bias both in the moment and phase determination. We saw no direct evidence that a precursory event could have been responsible for the additional energy release, as was suggested by Ihmle et al. (1993) ¿ American Geophysical Union 1994