Studying the deep-time origins of macroecological phenomena can help us to understand their long-term drivers. Given the considerable spatiotemporal bias of the terrestrial fossil record, it behooves us to understand how much biological information is lost. The aim of this study is to establish whether latitudinal diversity gradients are detectable in a biased terrestrial fossil record. I develop a simulated fossilization approach, weighting the probability of terrestrial mammal species appearing in the fossil record based on body size and geographic-range size; larger species with larger range sizes are more likely to enter the fossil record. I create simulated fossil localities from the modern North American mammal record. I vary the percentage of species successfully fossilized and estimate the magnitude of the latitudinal diversity gradient (slope of the richness gradient and degree of species turnover). I find that estimates of the latitudinal diversity gradient are sensitive to the loss of species with small body size and geographic-range sizes. In some cases, simulated fossil-record bias completely obliterates evidence of declining richness with latitude, a phenomenon that is not ameliorated by the application of nonparametric richness estimation. However, if the rate of preservation is medium (50% of species) to high (75% of species), the magnitude of the latitudinal diversity gradient can be reliably estimated. Similarly, changes in the diversity gradient estimates are largely explained by differences in the diversity–climate relationship among iterations, suggesting that these relationships may be measurable in the fossil record.