A new data set of over 43,000 surface marine air temperature (MAT) records covering the years 1795--1843 is described and a new methodology proposed for the homogenization of historical MAT data. The data consist of observations made at 1200 LT, as this was overwhelmingly the most common observation time in use before 1850. Simultaneous sea surface temperature (SST) is present in only 3.6% of records. For this reason, extension of historical marine temperature fluctuations back to the early 19th century will need to rely on MAT data. Nonstandard thermometer exposures and observing practices introduce nonclimatic biases into the data. Here I propose new correction schemes to account for the major sources of bias in the data. Similar principles are applicable to post-1850 data as well. The seeming disadvantage of nonstandard thermometer exposures and local noon observation time actually provides information that is used to adjust and homogenize the data. Unscreened thermometers absorb seasonally and geographically varying short wave radiation and give daytime temperatures that are too high. Calculation of grid box mean temperature anomalies, ΔTa, using a modern nighttime climatology highlights these biases. Ship-to-ship variations in thermometer calibration and performance are measured and adjusted using tropical ΔTa as a reference. Individual ships have large biases (in excess of 1 ¿C) which can dominate observed patterns of temperature anomalies in individual years. Using data from low latitudes, a regression equation is produced that predicts ΔTa as a function of net radiation. Predicted ΔTa at higher latitudes is then compared to observed values of ΔTa in the historical record to detect biases arising from internally generated ship's heat. The advantage to this approach is that SST data are not needed to correct MAT data, thereby preserving both as independent marine temperature records. Finally, a correction for nonstandard thermometer exposure is made. This last correction is made by comparing historical and modern noon minus nighttime temperature differences. The difference between the two is considered to be due to nonstandard thermometer exposure in the historical data. Since the local MAT diurnal range is assumed to be independent of secular changes in temperature the historical adjusted nighttime marine air temperature can be compared to modern climatologies. ¿ 2000 American Geophysical Union