The early Eocene (49-55 million years ago) is a time interval characterized by elevated surface temperatures and atmospheric CO 2 (refs 1,2), and a flatter-than-present latitudinal surface temperature gradient 3,4 . The multi-proxy-derived flat temperature gradient has been a challenge to reproduce in model simulations [5][6][7] 10 , invalidating the apparent, extremely warm polar sea surface temperatures. We conclude that there is a need to reinterpret TEX H 86 -inferred marine temperature records in the literature, especially for reconstructions of past warm climates that rely heavily on this proxy as reflecting subsurface ocean.The combination of global warming and high atmospheric CO 2 during the early Eocene renders this time interval a potential analogue for anthropogenic climate change. A long-standing, unresolved issue in simulating warmer-than-present climates is the failure of models in reproducing the flat Equator-to-pole surface temperature gradient often observed in proxy data 5,6 . The model-proxy match is improved in the terrestrial realm in high-CO 2 scenarios 6 or via tuning model parameters 7 , but the extreme surface oceanic warmth in polar regions inferred from TEX H 86 has been irreconcilable. Although this discrepancy was attributed to missing processes in climate models 8 , inconsistencies in proxy data also suggest deficiencies in proxy understanding 11,12 To investigate this hypothesis, we examine 22 pairs of U K 37 -and TEX H 86 -inferred temperature records (in total 5,528 samples) measured in tandem on sediment cores from sites spanning diverse oceanographic settings and timescales from thousands to millions of years ( Supplementary Figs 1 and 2). Temporal leads and lags between proxy record pairs are possible, as these proxies might reflect different water depths, where the timing of temperature changes differ 20 , and proxy-specific sedimentary processes 21 can create temporal offsets. This inhibits a direct comparison of the time series, as timing differences generally destroy coherency.Instead, power spectra of each of the paired U K 37 and TEX H 86 records are estimated and the mean power spectral estimate (PSD) for each proxy type is compared. Spectra averaged over multiple regions attenuate local effects, thereby facilitating inter-comparison between proxy types. This technique is insensitive to temporal offsets between the records 22 and allows fingerprinting the reasons for discrepancies between the proxies 23 . The power spectra of U K 37 and TEX H 86 show a very similar shape, with increased energy towards lower frequencies (Fig.