The effects of three selection strategies for dry cow therapy on prevention of new infections and rate of antibiotic usage were compared. Quarter infection status of 1044 cows in 12 herds was determined by bacteriological methods at drying off, calving and three to five months into the following lactation. Cows that were uninfected at drying off were randomly allocated to treatment (whole udder, dry cow therapy) and non-treatment groups. Infected cows were randomly allocated to whole udder or infected quarter only treatments. The strategies compared were blanket treatment (treat all quarters of all cows), selective cow treatment (treat all quarters of any cow infected in one or more quarters) and selective quarter treatment (treat infected quarters only). Selective cow treatment was identified as the preferred strategy. Blanket treatment resulted in increased antibiotic usage (15.5 vs 6.4 tubes per infection eliminated) with no additional benefit, and selective quarter treatment resulted in a higher new infection rate (6.4% vs 3.9% quarters) in the dry period. The prevalence of infection within a herd at drying off had no influence on new infection rates in the dry period or early lactation. The cure rate after dry cow treatment (mean of 66%) decreased significantly with increasing age (P < 0.001). Cows infected in the previous lactation contributed over 76% of infections at calving and nearly 70% at mid-lactation. To lower the incidence of mastitis in a herd, a greater emphasis on culling of older infected cows and prevention of new infections during lactation is needed.