Understanding how predators select their prey can provide important insights into community structure and dynamics. However, the suite of prey species available to a predator is often spatially and temporally variable. As a result, species-specific selectivity data are of limited use for predicting novel predator-prey interactions because they are assemblage specific. We present a method for predicting diet selection that is applicable across prey assemblages, based on identifying general morphological and behavioural traits of prey that confer vulnerability to predation independent of species identity. We apply this trait-based approach to examining prey selection by Indo-Pacific lionfish (Pterois volitans and Pterois miles), invasive predators that prey upon species-rich reef fish communities and are rapidly spreading across the western Atlantic. We first generate hypotheses about morphological and behavioural traits recurring across fish species that could facilitate or deter predation by lionfish. Constructing generalized linear mixed-effects models that account for relatedness among prey taxa, we test whether these traits predict patterns of diet selection by lionfish within two independent data sets collected at different spatial scales: (i) in situ visual observations of prey consumption and availability for individual lionfish and (ii) comparisons of prey abundance in lionfish stomach contents to availability on invaded reefs at large. Both analyses reveal that a number of traits predicted to affect vulnerability to predation, including body size, body shape, position in the water column and aggregation behaviour, are important determinants of diet selection by lionfish. Small, shallow-bodied, solitary fishes found resting on or just above reefs are the most vulnerable. Fishes that exhibit parasite cleaning behaviour experience a significantly lower risk of predation than non-cleaning fishes, and fishes that are nocturnally active are at significantly greater risk. Together, vulnerable traits heighten the risk of predation by a factor of nearly 200. Our study reveals that a trait-based approach yields insights into predator-prey interactions that are robust across prey assemblages. Importantly, in situ observations of selection yield similar results to broadscale comparisons of prey use and availability, which are more typically gathered for predator species. A trait-based approach could therefore be of use across predator species and ecosystems to predict the outcomes of changing predator-prey interactions on community dynamics.