Roadkill estimates can be heavily biased due to variable carcass persistence time. If not adequately considered, this bias may lead to incorrect management actions. We designed an experiment aimed to provide an accurate assessment of carcass persistence time according to road type and position on the road along 114 km of roads (24 in dirt roads, 74 two-lane roads, and 16 in four-lane roads). We used two types of rodent of different body size to perform the experiment: mice (30 ± 8 g) were placed on the road shoulders and road lanes every 1000 m, and rats (400 ± 55 g) were placed only on road shoulders every 2000 m. Carcasses were monitored for up to five consecutive days. The persistence times of mice and rats were in general similar, with an estimated median time of 1 day. However, we found considerable differences according to road type and position on the road: the estimated median persistence time was substantially longer for carcasses on the four-lane roads, and longer on the road shoulders compared to the traffic lanes. Overall, we estimated an annual mortality of small animals to be higher than 100.000 individuals in Federal District, of which ca. 10% in dirt roads. Our results confirm that a great proportion of carcasses are likely to be undetected in roadkill surveys, particularly small body sized animals, and that such errors are also related to the type of road and the position of carcasses on the road. We highlight that the impact of dirt and two-lane roads on wildlife may be greatly underestimated.