Acoustic-trawl (AT) survey methods are widely used to estimate the abundance and distribution of pelagic organisms. This technique relies on estimates of size and species composition from trawl catches along with estimates of the acoustic properties of these animals to convert measurements of acoustic backscatter into animal abundance. However, trawls are selective samplers, and if the catch does not represent the size and species composition of the animals in the acoustic beam the resulting abundance estimates will be biased. We conducted an experiment to quantify trawl selectivity for species encountered during an AT survey of the Alaska Arctic. The pelagic assemblage in this environment was dominated by small young-of-the-year (age-0) fishes and jellyfish, which may be poorly retained in trawls. A large midwater trawl (Cantrawl) and a smaller midwater trawl (modified Marinovich) were used during the survey. The Marinovich was equipped with 8 small-mesh recapture nets which were used to estimate the probability that an individual that enters the trawl is retained. In addition, paired hauls were made with the Cantrawl and Marinovich to estimate the difference in selectivity between the two trawls. A statistical model was developed to combine the catches of the recapture nets and the paired hauls to estimate the length-dependent selectivity of the trawls for the most abundant species (e.g., age-0 fishes and jellyfish). The analysis indicated that there was substantial size and species selectivity: although the modified Marinovich generally had a higher catch per unit effort, many of the animals encountered in this environment were poorly retained by both trawls. The observed size and species selectivity of the trawls can be used to select appropriate nets for sampling pelagic fishes, and correct survey estimates for the biases introduced in the trawl capture process.