Branched glycerol dialkyl glycerol tetraethers (brGDGTs) are bacterial membrane lipids that have been widely used as tools for the quantitative reconstruction of past terrestrial temperatures in lakes. Two approaches, i.e., regional/global and site-specific calibrations, have been previously employed in the reconstruction. However, their strengths and weaknesses have not been properly evaluated yet, making it challenging to choose a suitable approach to paleoreconstruction. Here, we investigated the distribution of brGDGTs in surface sediments of 65 lakes along a mean annual lake water temperature (MLWT) transect in West China and developed a regional MBT'_5ME (methylation degree of 5-methylated brGDGTs) calibration for MLWT reconstruction. Meanwhile, a site-specific brGDGT calibration for Lake Yangzonghai (YZH), located within the West China transect, was generated based on the relationship between the instrumental temperature record over the past 100 years and individual brGDGT abundances in a short sediment core taken from the lake. These two calibrations were applied to global lake surface sediments, as well as a long sediment core spanning the Holocene from Lake YZH. The linear correlations between MBT'_5ME and temperature are highly consistent among lakes from different regions, indicating the near-universal response of brGDGTs in global lakes to temperature. However, MLWT inferred from the site-specific calibration shows a weak relationship with actual MLWT in global lakes, indicating that the site-specific calibration might be only a result of overfitting rather than reflecting the response of brGDGT distribution to ambient temperature. Applications to a Holocene sediment core in Lake YZH suggest that the regional MBT'_5ME calibration is more suitable for reconstructing paleotemperature than the site-specific calibration. The regional MBT'_5ME calibration derived temperature record shows a cold Younger Dryas (YD), followed by a rapid warming of ∼3 °C during the last deglaciation, and a long-term cooling of ∼1 °C from 9.9 to ∼0.2 cal ka BP. The principal component analysis shows that temperature is the dominant factor affecting brGDGT distributions in the long sediment core whereas the site-specific calibration is strongly affected by human activities rather than reflecting the ambient temperature. We highlight the importance of using appropriate approaches to developing calibrations for paleotemperature reconstruction and assessing the reliability of the temperature records.