Knowledge of past environmental changes is largely derived from geochemical proxies. Commonly used proxies to reconstruct paleo seawater temperature include UK’37 and TEX86, which are based on the lipids of haptophyte and marine archaea, respectively. Proxies often suggest divergent amplitudes and temporal trends in the temperature evolution at a given site. These divergences are usually explained by invoking differences in the seasonal production and the habitat depth of the source organism of proxies. In order to benefit from the proxy-based paleo climatic data to improve our understanding of future climate, a quantitative interpretation of the proxy data, including knowledge of their uncertainty, is needed. In this talk, I will present ongoing work on the reliability of UK’37 and TEX86. The work spans the uncertainty of marine sediment core-top calibrations, the validity of proxy-temperature relationships over time, and the sources of uncertainty in the proxy signal. It is generally assumed that the broader the geographical coverage and temperature range of a data set, the more robust the resulting proxy calibration is. We compiled our 160 new marine surface-sediment TEX86 data with various published data sets, resulting in a global calibration that is based on ~600 data points. Our calibration is similar to the previous global calibration, but with a larger standard error of estimate. The scatter in the newly expanded global data set is due to spatially varying proxy-temperature relationships. The finding raises the question whether a universal calibration should be preferred over regional calibrations in some oceanic regions. To test whether the modern spatial calibration can be used to quantitatively determine past temperatures, we compiled temperature records from global sites with multi-proxy data. Simulated glacial seasonal and vertical temperature distributions from state-of-the-art climate models are used as an independent means for appraising the seasonal and water depth differences in proxies. It is envisioned that the model-data comparison will shed light on any systematic spatial and temporal relationships between proxies. In order to reduce the uncertainty and make optimal use of the proxy data, we have to understand the sources of the uncertainty. Thus we measured UK’37 and TEX86 on several marine sediment cores of the same multicoring deployment (maximum distance between the cores < 1.5m). The findings from this exercise enable us to quantify potential errors in a proxy record, including errors caused by the instrument, the work-up procedure and the spatial heterogeneity of proxies in sediments. Knowledge of these uncertainties allows us to disentangle proxy noise from climate variability in a proxy record.