To minimise the error in biomass estimates from stratified random trawl surveys, the allocation of trawl stations amongst strata should be made according to prior information on fish distribution. Often such information is slight. A 2‐phase survey strategy is proposed in which allocation of trawl stations in Phase 2 is based on catches obtained in Phase 1. Computer simulations based on 2 sets of real survey data indicate that this strategy reduces the skew in the biomass estimate (thus reducing the probability of gross over‐estimation) as well as reducing the expected error. The efficiency of the 2‐phase design compared to the conventional design was 127% and 180% on the 2 survey areas. In additon, the adaptive nature of the design strategy allows for greater flexibility at sea.