Wildlife collisions with human-built structures are a major source of direct anthropogenic mortality. Understanding and mitigating the impact of anthropogenic collisions on wildlife populations require unbiased mortality estimates. However, counts of collision fatalities are underestimated due to several bias sources, including scavenger removal of carcasses between fatality surveys and imperfect detection of carcasses present during surveys. These biases remain particularly understudied for bird-window collisions, the largest source of avian collision mortality. In Stillwater, Oklahoma, USA, we used bird carcasses collected during window collision monitoring to experimentally assess factors influencing scavenging and observer detection, and we employed trail cameras to characterize the scavenger community and timing of scavenging. We recorded nine scavenger species, but the domestic cat and Virginia opossum were responsible for 73% of known-species scavenging events. The most frequent scavenger species were primarily nocturnal, and 68% of scavenging events occurred at night. Scavenger species best predicted time to first scavenging event, season best predicted carcass persistence time, and both season and carcass size predicted whether any carcass remains persisted after scavenging. Our results also suggest that observer detection was influenced by substrate, with greater detection of carcasses on artificial substrates. Our findings related to scavenging timing have important implications for the unbiased estimation of collision mortality because the timing of peak scavenging relative to timing of peak mortality can substantially influence accuracy of adjusted mortality estimates. Further, the differences in correlates for time to first scavenging and time to carcass removal (i.e., persistence time) illustrate the importance of explicitly measuring these often-independent events that are frequently conflated in the anthropogenic mortality literature.