The cost of large-scale association studies may be reduced substantially by analysis of pooled DNA from multiple individuals. Here we examine the optimal symmetric and asymmetric designs for pooling experiments for quantitative traits under a range of assumptions about the underlying genetic model and the sources of experimental errors in allele frequency estimation. The results indicate that, in the absence of experimental errors and for common alleles with additive effects, a symmetric pooling scheme comparing the top 27% with the bottom 27% of the trait distribution is optimal, extracting 80% the total information available. A symmetric design is not optimal for rare or recessive alleles, which require asymmetric (or other) pooling strategies. Allele frequency measurement errors reduce the optimal pooling fraction as well as the overall efficiency of the pooling design. In contrast, random variation in the amount of DNA contributed by individuals to a pool reduces only the overall efficiency of the pooling design. Our results emphasize the importance of minimising experimental errors and suggest a pooling fraction of around 20%.