The El Niño-Southern Oscillation (ENSO) influences climate and weather over much of the globe, so uncertainties in its response to external forcing 1 hinder long-range climate predictability. Modeling studies have argued that such forcings may affect ENSO either via the seasonal cycle [2][3][4][5][6] or extratropical influences [7][8][9] , but this link is poorly constrained by the short instrumental record. Here we synthesize a pan-Pacific network of high-resolution marine biocarbonates spanning discrete snapshots of the Holocene (past ∼ 10,000 years), which we use to constrain a set of global climate model (GCM) simulations via a forward model 10 and a consistent treatment of uncertainty. Observations suggest important reductions in ENSO variability throughout the interval, most consistently during 3-5 kyBP, when ∼2/3 reductions are inferred. The magnitude and timing of these ENSO variance reductions bear little resemblance to those simulated by GCMs, or to equatorial insolation. The central Pacific witnessed a notable mid-Holocene increase in seasonality, at odds with the reductions 11 simulated by GCMs. Finally, while GCM aggregate behavior is consistent with an inverse relationship between seasonal amplitude and ENSO-band variance in sea-surface temperature 3,6,12, 13 , this relationship is not borne out by these observations. The synthesis suggests that tropical Pacific climate is highly variable, but exhibited millennia-long periods of reduced ENSO variability whose origins, whether forced or unforced, are a crucial issue for model development and long-term climate prediction.2 ENSO, the non-linear interaction between the tropical Pacific atmosphere and ocean, is the leading pattern of global interannual variability, with important physical, ecological, and human impacts. Yet, predicting its long-term behavior in the face of continued greenhouse forcing has proven elusive 1 . While the predictive skill of climate models at interannual timescales can be tested using instrumental observations, such records are too short to evaluate the fidelity of modelsimulated tropical Pacific variability on decadal-to centennial-timescales, i.e. those relevant for future climate projections. This motivates the use of paleoclimate observations, which cover a much longer time span and predate the observations used to develop and tune climate models, hence providing an out-of-sample test of their predictive ability 14 .The mid-Holocene (MH, ca 6,000 yrs before present; 6 kyBP) represents a key target for evaluating the simulated response of ENSO to changes in external forcing. While ice volume and greenhouse gas concentrations were essentially similar to today, the latitudinal and seasonal distribution of incoming solar radiation (insolation) was markedly different as a result of precession 15 : seasonal contrast was amplified in the northern hemisphere and reduced in the southern hemisphere. Thus, the mid-Holocene provides an opportunity to explore the link between changes in the seasonal cycle, meridional asymmetry in th...
