Across the world, many wildlife studies rely on data collected by volunteers. Roadkill studies often rely on data collected by non-experts including road maintenance personnel and volunteers, but data quality control is rarely applied. We investigated whether maintenance personnel correctly identified the species of road-killed mammals along toll roads in São Paulo State, Brazil. We investigated 3222 images of road-killed animals and compared the original species descriptions by road maintenance personnel (non-experts) with our identification (experts). We also presented images of alive and road-killed mammals to road maintenance personnel (n = 179) and asked them to describe the species. We found that road maintenance personnel typically correctly identified certain common, large, or highly recognizable species. However, rare or rarely seen species, species that resemble other species (e.g. small wild canids and felids), or species that are not highly recognizable were often misidentified, ambiguously described, or not identified at all. We also found that the ability of road maintenance personnel to correctly identify the most common road-killed small wild canids and felids is dependent on the context. When similar species are rare, road maintenance personnel typically correctly identifies the most common road-killed small wild canids and felids. However, common small canids and felids are not reliably identified if similar species are more abundant. To improve the reliability of species identification by non-experts, we recommend training in species identification, including images with a scale to accompany all roadkill records, and verification of the roadkill records and associated images for selected species by experts.