The paleothermometer based on isoprenoid glycerol dialkyl glycerol tetraethers (isoGDGTs) has widely been applied in various marine settings to reconstruct past sea surface temperatures (SSTs). However, it still remains uncertain how well this proxy reconstructs annual mean SSTs. Here, we assess environmental factors governing the paleothermometer in the Mediterranean Sea, by studying the distribution of isoGDGTs in surface sediments, suspended particulate matter (SPM), and two sediment cores. A redundancy analysis using the fractional abundance of the six major isoGDGTs indicates that the sedimentary isoGDGTs are mostly influenced by three environmental factors explaining a large part (74%) of the variance in isoGDGT distribution. In order of decreasing significance, these factors are annual mean SST, continental organic matter input as indicated by the BIT index, and water depth. However, when considering only the four isoGDGTs that are used for the proxy, water depth is the most significant parameter, explaining 63% of the variance. Indeed, a strong positive relationship between water depth and is observed in both surface sediments and SPM from the Mediterranean Sea. This is driven by an increase in fractional abundances of GDGT-2 and crenarchaeol regio-isomer and a decrease in the fractional abundances of GDGT-1 and GDGT-3 with increasing water depth, leading to a bias to higher temperatures of in deep-water surface sediments. The fact that the water-depth trend is also apparent in SPM suggests that this change might be due to a change in thaumarchaeotal community thriving below surface mixed-layer waters and that this signal is, at least partly, incorporated into sedimentary isoGDGTs. Interestingly, surface-sediment values from >1000 m water depth do not show a correlation with water depth anymore and instead are correlated to annual mean SSTs. A composite deep-water dataset of surface sediments from both the Mediterranean Sea and the Red Sea, interconnected regional restricted basins with relatively high bottom-water temperatures and high salinity, forms a distinctive correlation line, statistically distinct from that of the general global correlation. Application of this correlation on two sedimentary records from the western Mediterranean Sea covering the last deglaciation yields SSTs nearly identical to those obtained with the paleothermometer, whereas the global calibration substantially overestimates SSTs. Our results show that the warm bias of the proxy in the Mediterranean Sea is not due to seasonality, as previously suggested. Further research is needed to elucidate the mechanism behind the strong water depth trend of in the Mediterranean Sea which is not apparent in open ocean settings.