We used female ring-necked pheasant (Phasianus colchicus) carcasses as surrogates for greater sage-grouse (Centrocercus urophasianus) to study factors influencing survival and detection bias associated with avian fence collision surveys in southern Idaho, USA, during spring 2009. We randomly placed 50 pheasant carcasses on each of 2 study areas, estimated detection probability during fence-line surveys, and monitored survival and retention of carcasses and their associated sign over a 31-day period. Survival modeling suggested site and habitat features had little impact on carcass survival, and constant survival models were most supported by the data. Model averaged carcass daily survival probability was low on both study areas and ranged from 0.776 to 0.812. Survival of all carcass sign varied strongly by location, and the top sign survival model included a site effect parameter. Model averaged daily survival probability for collision sign on the 2 study sites ranged from 0.863 to 0.988 and varied between sites. Logistic regression modeling indicated detection probability of carcasses during fence-line surveys for avian collision victims was influenced by habitat type and microsite shrub height at the carcass location. Carcasses located in big sagebrush (Artemisia tridentata) habitats were detected at a lower rate (0.36) than carcasses in little (A. arbuscula) and black sagebrush (A. nova) habitats (0.71). Increasing shrub height at the carcass location from the little sagebrush mean of 16.5 cm to the big sagebrush mean of 36.0 cm reduced detection probability by approximately 30%. Avian fence collision surveys in sagebrush-steppe habitats should be conducted at ≤2-week sampling intervals to reduce the impact of survival bias on collision rate estimates. Two-week sampling intervals may be too long in areas with low carcass and sign survival, therefore survival rates should be estimated on all study areas to determine the appropriate sampling interval duration. Researchers should be aware of the effects of local vegetation on detection probabilities, and methods to correct detection probabilities based on collision site attributes should be applied to ensure more accurate collision rate estimates. Additionally, caution should be used when aggregating or comparing uncorrected collision data from areas with differing vegetation, as detection probabilities are likely different between sites.