Forest autotrophic respiration (R(a)) plays an important role in the carbon balance of forest ecosystems. However, its drivers at the global scale are not well known. Based on a global forest database, we explore the relationships of annual R(a) with mean annual temperature (MAT) and biotic factors including net primary productivity (NPP), total biomass, stand age, mean tree height, and maximum leaf area index (LAI). The results show that the spatial patterns of forest annual R(a) at the global scale are largely controlled by temperature. R(a) is composed of growth (R(g)) and maintenance respiration (R(m)). We used a modified Arrhenius equation to express the relationship between R(a) and MAT. This relationship was calibrated with our data and shows that a 10 degrees C increase in MAT will result in an increase of annual R(m) by a factor of 1.9-2.5 (Q10). We also found that the fraction of total assimilation (gross primary production, GPP) used in R(a) is lowest in the temperate regions characterized by a MAT of approximately 11 degrees C. Although we could not confirm a relationship between the ratio of R(a) to GPP and age across all forest sites, the R(a) to GPP ratio tends to significantly increase in response to increasing age for sites with MAT between 8 degrees and 12 degrees C. At the plant scale, direct up-scaled R(a) estimates were found to increase as a power function with forest total biomass; however, the coefficient of the power function (0.2) was much smaller than that expected from previous studies (0.75 or 1). At the ecosystem scale, R(a) estimates based on both GPP - NPP and TER - R(h) (total ecosystem respiration - heterotrophic respiration) were not significantly correlated with forest total biomass (P > 0.05) with either a linear or a power function, implying that the previous individual-based metabolic theory may be not suitable for the application at ecosystem scale.