Prevention trials using incidence or mortality as endpoints require a large number of participants and long follow-up. Trials using biomarkers as endpoints would potentially require fewer participants, less time, and significantly less resources to test promising new agents for breast cancer prevention. To test this idea, a randomized trial of tamoxifen for 1 year versus observation for 1 year was conducted to determine whether tamoxifen can cause regression of hyperplastic breast tissue, whether it changes the biomarker phenotype of premalignant disease or normal breast epithelium, and if biomarkers can be used as early surrogate indicators of response to tamoxifen. Women were identified by having an abnormal mammogram and ductal hyperplasia diagnosed by core needle biopsy. Image-directed needle biopsy was repeated in the same site of the breast after 1 year. Approximately 3000 women were screened, and 265 were eligible. Sixty-three women were randomized and paired biopsies from 45 subjects were available for analysis. There was no evidence of substantial regression of hyperplasia--fewer samples showed hyperplasia at 1 year follow-up, but this was seen in both untreated and tamoxifen-treated women. There were trends for reductions in ER and PgR and trends for increases in bcl-2 in normal and hyperplastic tissue in the tamoxifen-treated arm, though these changes did not reach statistical significance. Proliferation, determined by Ki67 staining, was not significantly changed. Clinical trials of this type are difficult to carry out and modifications in trial design are needed to make this process more efficient.