We present a detailed comparison of five “transfer function” techniques calibrated to reconstruct sea-surface temperature (SST) from planktic foraminifer counts in western Pacific surface sediments. The techniques include the Imbrie–Kipp method (IKM), modern analog technique (MAT), modern analog technique with similarity index (SIMMAX), revised analog method (RAM), and the artificial neural network technique (ANN). The calibration is based on a new database of 694 census counts of planktic foraminifers in coretop samples from the western Pacific, compiled under a cooperative effort within the MARGO (multiproxy approach for the reconstruction of the glacial ocean surface) project. All five techniques were used to reconstruct SST variation in a well-dated Holocene to last glacial maximum interval in core MD972151 from the southern South China Sea (SCS) to evaluate the magnitude of cooling in the western tropical Pacific during the LGM. Our results suggest that MAT, SIMMAX, RAM and ANN show a similar level of performance in SST estimation and produce ⩽1°C uncertainties in coretop SST calibrations of the western Pacific. When applying these techniques to the downcore faunal record, the IKM, which performed significantly worst in the calibration exercise, produced glacial SST estimates similar to present-day values, whereas the other four techniques all indicated ∼1°C cooler glacial SST. Because of their better performance in the calibration dataset, and because of the convergence among the techniques in the estimated magnitude of glacial cooling in the studied core, we conclude that MAT, SIMMAX, RAM and ANN provide more robust planktic foraminifer paleo-SST estimates than traditional IKM techniques in western Pacific paleoceanographic studies.