In 2002 a study reported that a blood test, based on pattern-recognition proteomics mass spectroscopy analysis of serum, was nearly 100% sensitive and specific to detect ovarian cancer. Plans to introduce a commercial screening test by early 2004 were delayed amid concerns about whether the approach was reproducible and reliable. In this issue of JNCI, two commentaries discuss whether the initial results are reproducible and whether bias may account for results. This essay describes how threats to validity from chance and bias may cause erroneous results and inflated expectations in the kind of observational research being conducted in several “-omics” fields to assess molecular markers for diagnosis and prognosis of cancer. To address such threats and to realize the potential of new -omics technology will require application of appropriate rules of evidence in the design, conduct, and interpretation of clinical research about molecular markers.