Significant discrepancies exist between the detrended variability of late-Holocene marine temperatures inferred from Mg/Ca and Uk37 proxies, with the former showing substantially more centennial-scale variation than the latter. Discrepancies exceed that attributable to differences in location and persist across various calibrations, indicating that they are intrinsic to the proxy measurement. We demonstrate that these discrepancies can be reconciled using a statistical model that accounts for the effects of bioturbation, sampling and measurement noise, and aliasing of seasonal variability. The smaller number of individual samples incorporated into Mg/Ca measurements relative to Uk37 measurements leads to greater aliasing and generally accounts for the differences in the magnitude and distribution of variability. An inverse application of the statistical model is also developed and applied in order to estimate the spectrum of marine temperature variability after correcting for proxy distortions. The correction method is tested on surrogate data and shown to reliably estimate the spectrum of temperature variance when using high-resolution records. Applying this inverse method to the actual Mg/Ca and Uk37 data results in estimates of the spectrum of temperature variance that are consistent. This approach provides a basis by which to accurately estimate the distribution of intrinsic marine temperature variability from marine proxy records. © 2013 Elsevier B.V.