Climatic and oceanographic processes in the Pacific Ocean have global implications. Reliable sedimentary proxies in this region are thus necessary to understand the magnitude of past climate variability. Radiolarian assemblages are of particular relevance in some regions of the Pacific Ocean where other proxies are either poorly preserved or are biased towards a specific season. Previously published radiolarian-based transfer functions in the Pacific Ocean did not cover some regions, such as the Western Pacific marginal seas or high-latitude Southwestern Pacific. Filling these gaps in sample distribution and using a thorough methodology for finding the most important variable for radiolarian assemblages in surface sediment are both essential steps in quantitative studies. In this study, we use published radiolarian census counts from 801 core-tops distributed across the Pacific Ocean, to investigate which environmental factors drive the assemblages. We harmonized taxonomically the modern microfossil dataset, and using multivariate statistical analyses, we determined that sea surface temperature at 10 m (SST10) is the most important variable influencing the changes observed in the radiolarian assemblages. The calibration method weighted-mean modern analogue technique with five analogs (WMAT-K5) corrected for autocorrelation using a cut-off distance of 500 km, showed a performance of R2 cv = 0.83; RMSEP = 3.8 °C. This calibration method was then applied to taxonomically harmonized radiolarian census counts from 31 cores located in the Bering Sea, Western Pacific marginal seas, and Southwest and Eastern Equatorial Pacific, with some of these records covering at least the last 165 ka. We assessed the analogue quality and significance of the downcore SST10 reconstructions in all of them. We found that temperatures at 10 and 200 m were the most significant variables for the fossil assemblages. Finally, we compare the temperature reconstructions to previously published radiolarian-based SST estimates for the same cores, or to other SST records based on other methodologies. We find some differences between our new temperature estimates compared to existing ones, in particular in the Eastern Equatorial Pacific. Comparison to other methods in the Japan and Bering Sea and Sea of Okhotsk show that radiolarian-based reconstructions provide robust temperature estimates compared to biogeochemical methods, which showed SST overestimation during glacial periods. • New radiolarian calibration dataset with 801 core-tops samples in the Pacific Ocean. • Sea surface temperature at 10 m is the main driver for the radiolarian assemblages. • Application of the radiolarian transfer function in 31 downcore radiolarian records. • In high latitude regions, sub-surface temperatures at 200 m are also significant. • Some regions show diverging radiolarian-based temperatures compared to alkenones. [ABSTRACT FROM AUTHOR]