Nearly ten years of daily rawinsonde data for Canton Island (3S, 172W) have been subjected to spectrum and cross-spectrum analysis. In the course of this analysis a very pronounced maximum was noted in the co-spectrum of the 850- and 150-mb zonal wind components in the frequency range 0.0245–0.0190 day−1 (41–53 days period). Application of a posteriori sampling theory resulted in a significance level of ∼6% (0.1% prior confidence level). This type of significance test is appropriate because no prior evidence or reason existed for expecting such a spectral feature. Subsequent analysis revealed the following structure of the oscillation. Peaks in the variance spectra of the zonal wind are strong in the low troposphere, are weak or non-existent in the 700–400 mb layer, and are strong again in the upper troposphere. No evidence of this feature could be found above 80 mb, or in any of the spectra of the meridional component. The spectrum of station pressure possesses a peak in this frequency range and the oscillation is in phase with the low tropospheric zonal wind oscillation, and out of phase with that in the upper troposphere. The tropospheric temperatures exhibit a similar peak and are highly coherent with the station pressure oscillation; positive station pressure anomalies are associated with negative temperature anomalies throughout the troposphere. Thus, the lower-middle troposphere appears to be a nodal surface with u and P oscillating in phase but 180° out of phase above and below this surface. Evidence for this phenomenon was found in shorter records at Kwajalein (9N, 168E) but not at Singapore (1N, 104E) or Balboa, Canal Zone (9N, 79w). We speculate that the oscillation is a large circulation cell oriented in zonal planes and centered in the mid-Pacific.