A homogeneous, consistent, high-quality in situ temperature dataset covering some decades in time is crucial for the detection of climate changes in the ocean. For the period from 1940 to the present, this study investigates the data quality of temperature profiles from mechanical bathythermographs (MBT) by comparing these data with reference data obtained from Nansen bottle casts and conductivity–temperature–depth (CTD) profilers. This comparison reveals significant systematic errors in MBT measurements. The MBT bias is as large as 0.2°C before 1980 on the global average and reduces to less than 0.1°C after 1980. A new empirical correction scheme for MBT data is derived, where the MBT correction is country, depth, and time dependent. Comparison of the new MBT correction scheme with three schemes proposed earlier in the literature suggests a better performance of the new schemes. The reduction of the biases increases the homogeneity of the global ocean database being mostly important for climate change–related studies, such as the improved estimation of the ocean heat content changes.