The calculation of global land surface air temperature trends using the instrumental record has been based primarily upon two methods of maximizing the availability of station records. Hansen and Lebedeff[l981] developed a technique that is still used today, known as the reference station method; Jones et al. [1986a] popularized the climate anomaly method in their calculations of global temperature trends. In this paper we introduce yet another approach designed to maximize station records, referred to as the first difference method. To test the sensitivity of global temperature trend analysis to the method used, we calculate worldwide-averaged land surface mean temperature using each of these methods with an identical data base, the Global Historical Climatology Network. For further comparisons, a global climate model (GCM) transient model simulation is interpolated to the Global Historical Climatology Network station locations and the three techniques are then applied to data interpolated to the station locations from the model. The Intergovernmental Panel on Climate Change (IPCC); [Nicholls et al. 1996] estimated a global land and ocean temperature change of 0.45°C±0.15°C since the 19th century. Their assessment of the uncertainty associated with this temperature trend did not specifically address the differences that the method of calculating a global temperature time series might produce. Our results indicate that the differences in 1880–1990 trends produced by these three different methods are only a few hundredths of a degree centigrade per 100 years on trends of approximately 0.5°C/100 years. This is quite small compared to the 0.15°C/100 years uncertainty associated with the IPCC global land and ocean assessment which included factors such as data homogeneity which are not addressed here. Indeed, our results indicate that the source of differences in trends is more likely to be the method used to calculate a linear trend from a global temperature time series than the method used to create the global temperature time series. The modeled results confirm this finding but highlight other important characteristics: the reference station method has uncharacteristically low interannual variance, more similar to time series from the entire globe (land and ocean) than the global land area from which the data were observed. This lower variance can impact the statistical significance associated with linear trends.