The evolution of mimicry, and particularly the persistence of undefended Batesian mimetic forms that are imperfect copies of their defended models, remains a central question in evolutionary biology. Previous work has demonstrated that variation in mimetic fidelity in artificial prey can alter survival. However, no studies have validated the assumption that detailed laboratory-based measurements of mimetic fidelity are actually reflected in survival in natural field experiments. Here, we demonstrate that, in line with previous studies, the mimetic similarity of 77 hover fly (Diptera: Syrphidae) species to the common wasp Vespula alascensis is strongly related to the number of abdominal stripes exhibited by the flies. We then produce three artificial pastry baits: (i) a "model" which is chemically defended and has two stripes, (ii) a one-stripe mimic, and (iii) an unstriped mimic. Based on the ratings study, we predicted that the one-stripe mimic would exhibit survival intermediate between the unstriped mimic and the model. Baits were deployed in experiments each involving 81 baits (27 of each kind), at 3 sites, with experiments replicated 10 times at each site for a total deployment of 2430 baits. Proportional hazards models show that both one-striped and model baits survived equally well and significantly better than the unstriped baits, suggesting categorical prey identification rather than the use of stripe number as a continuous trait, as was suggested by the laboratory study. These findings suggests that, while humans and avian predators can distinguish mimics and models in the laboratory using a range of traits, behaviour in the field may not reflect this ability. This absence of a link between continuous measures of mimetic fidelity and prey selection may contribute to the maintenance of imperfect mimicry, but more studies using near-natural experimental paradigms are needed to investigate the phenomenon further.