Although standards of practice exist and are widely accepted for evaluating new treatment regimens, such do not exist for evaluating new biomarkers. This talk will draw on experiences with research on cancer biomarkers for diagnosis, screening, risk prediction and prognosis, conducted through the NCI sponsored Early Detection Research Network. We will discuss the 4 broad areas that must be considered in designing a definitive biomarker validation study: (i) the clinical context/populations in which the biomarker will be used; (ii) performance criteria that the biomarker must meet; (iii) the biomarker test itself; and (iv) the statistical power to draw conclusions. Key issues in each of these areas will be discussed both in general and in the context of two specific cancer biomarker studies. The prospective-specimen-collection-blinded-evaluation (PCBE) design will be formalized.