Several methods of low-frequency analysis are used to determine the source mechanism of the July 31, 1970, Great Colombia earthquake and to reexamine two controversial results previously obtained for this event. In a classic study of the earthquake, Dziewonski and Gilbert <1974> concluded (1) that low-frequency (f<4 mHz) mode spectrum observations require an isotropic compression at the source of a magnitude similar to the deviatoric moment release and (2) that the isotropic component of moment release precedes the short-period onset of the earthquake by 80 s. The original data set collected by Dziewonski and Gilbert is used in a reanalysis of the Great Colombia earthquake, taking advantage of the theoretical and computational advances that have been made during the past 20 years in predicting normal mode spectra. In particular, the splitting and coupling of modes induced by rotation, ellipticity, and three-dimensional (3-D) mantle structure are considered. When splitting and coupling are taken into account in the analysis of the observed spectra, the isotropic component is reduced to an insignificant size, and the data are well explained by a purely deviatoric source which does not require the initiation of moment release before the short-period onset time of the earthquake. When the effects of splitting and coupling are ignored in the calculation, the results are instead similar to those of Dziewonski and Gilbert <1974>. Experiments with synthetic spectra generated for the deviatoric part of the moment tensor confirm that the distortion of the low-frequency end of the normal mode spectrum caused by splitting and coupling is incorrectly interpreted as an isotropic component when these effects are ignored in the inverse problem. The detailed effects of modeling errors of this kind on the retrieved source parameters will, in general, depend both on the source mechanism and the distribution of observing stations.¿ 1997 American Geophysical Union