We address several issues regarding the Antarctic Circumpolar Wave (ACW). These include the extent to which the ACW is related to El Ni¿o-Southern Oscillation (ENSO) cycles and the need to reconcile differences between the observed zonal wavenumber 2 pattern and the wavenumber 3 pattern found in coupled ocean-atmosphere models. In the present study we revisit the sea surface temperature (SST) signal of the observed ACW. We find that the associated SST variations may be decomposed into several empirical orthogonal functions (EOFs) that are linearly independent. The first four EOFs describe 84% of the total variance on timescales of 3--7 years, and superposition of SST signals of the four EOFs reproduces SST signals of the observed ACW. Given that EOF analysis can only describe standing oscillations, we then employ complex EOF (CEOF) analysis suitable for studying propagating waves. The first two CEOFs (denoted CEOFSST1 and 2) account for 70% of the total variance, and the SST variability reconstructed from them again reproduces the observed ACW. The real part of CEOFSST1 (44% of the total variance) is similar to the first real EOF (EOFSST1). Both EOFSST1 and CEOFSST1 may be identified with ENSO and have a wavenumber 2 zonal pattern in the southern middle-to-high latitudes. CEOFSST1 is the major constituent of the ACW in the period of 1981--1997. The associated atmospheric circulation displays a Pacific-South America (PSA) pattern. Statistical analyses demonstrate that the PSA pattern teleconnected to ENSO is the major mechanism that forces this principal constituent of the ACW. CEOFSST2 (26% of the total variance) has a zonal wavenumber 3 pattern in the same latitude band, similar to that simulated by several coupled models. We suggest that if ENSO is not adequately resolved in coupled models but structures with zonal wavenumber 3 pattern are, then the modeled ACW would appear to have a zonal wavenumber 3 pattern. ¿ 2001 American Geophysical Union