The striking colour-pattern variation of some aposematic species is paradoxical because selection by predators is expected to favour signal uniformity. Although the mechanisms allowing for the maintenance of such variation are not well understood, possible explanations include both non-adaptive processes like drift and gene flow; and adaptive processes, such as an interaction between natural and sexual selection, spatial and temporal variation in selection, a link between behaviour or other fitness-related traits and phenotype, and predators’ ability to generalise among different signals. Here we test whether warning-signal polymorphisms, such as that of dyeing poison frogs (Dendrobates tinctorius), could be maintained by differences in detectability among morphs. We did experiments in the wild using wax models with different aposematic colour patterns vs. cryptic ones, and examined the attack rates by wild predators over time. We also tested the detectability of different aposematic morphs by ‘human predators’ under different light environments. We found that cryptic frog models were attacked more than aposematic models, but there were no differences in bird attack rates towards the different aposematic morphs. However, we found that detectability of different morphs depends both on predator experience and light environment. We suggest that the interaction between differential detectability and signal efficiency among morphs in different light conditions could be a mechanism aiding to the maintenance of warning-signal polymorphisms. Our results highlight the importance of considering the light environment at which predators have their first encounters with aposematic prey for future studies on predation in the wild.