Leaf margin analysis (LMA), which is based on a correlation between the proportion of woody dicot species with non-toothed leaf margins and mean annual temperature, has been promoted as a tool for estimating mean annual temperature (MAT) from fossil-leaf assemblages. The original LMA calibration was based on East Asian mesic vegetation, and substantially the same relationship has been shown for other geographical regions, including Australian mesic vegetation. In this report, taphonomic effects are assessed using autochthonous samples from extant Australian forests for sites ranging from tropical lowland rainforest and monsoonal deciduous woodland to temperate rainforest with and without emergent Eucalyptus, and for parautochthonous and allochthonous (i.e., streambed) leaf accumulations. MAT was estimated within the binomial sampling error of the estimate for 27 of 30 (90%) of the test sites, and was found to underestimate MAT systematically when applied to streambed leaf assemblages. This result may reflect the streamside bias detected in recent studies of tropical forests in South America. Sites where MAT was overestimated are of low species richness (<10 spp.).