Abundance estimates from bottom trawl surveys are important elements in the evaluation of stock status. The reliability of stock assessment results are greatly influenced by the precision of the survey abundance estimates. In this study, a new stratified random design, based upon historical spatial distribution, for estimating the abundance of ages 5-12 Atlantic cod Gadus morhua on the eastern Scotian Shelf off Nova Scotia was evaluated. Gains in precision from the new design, relative to simple random sampling, were mainly due to the improved allocation of trawl sets to strata, whereas the gains solely due to the new stratification scheme were minimal. We conclude that any reasonable stratification scheme that is flexible enough to allow for the optimization of the allocation of sets to strata should perform well. In our study, the allocation scheme was based upon the historical spatial distribution of abundances. Although annual changes in spatial distribution would require revision of the allocation strategy, we found that averaging results over several earlier years was more robust than just using the most recent year's data. The new design has been used for an annual March survey of this Atlantic cod stock from 1986 to the present. Comparisons of the results from the 1986-1990 period with those from 1979-1984, which stemmed from the traditional depth-based stratification, indicate that the new design is more efficient with respect to gain in precision than the old design. Finally, we note that the simple random sampling variance is smaller in a July series of surveys of this stock than in the March series and this difference can be exploited to give more precise stratified surveys in July than in March.