The hypothesis tested was that mutant tumor DNA shed into plasma would show predictive value for monitoring response to therapy in non-small cell lung carcinoma patients. Pretreatment plasma specimens from 25 patients on a phase I trial were evaluated, 12 with paired posttreatment specimens, and 138 patients on the Southwest Oncology Group S0003 trial, 38 with paired posttreatment specimens, for the presence of K-RAS mutations. Thirteen tumor specimens from the phase I trial patients and seven tumor specimens from S0003 patients were also available for comparative analysis of K-RAS mutations in tumor tissue and plasma. All patients were treated similarly with paclitaxel, and carboplatin chemotherapy. DNA was extracted and mutational analyses performed using an RFLP-PCR assay. K-RAS mutations were found in plasma DNA in 5/25 of the phase I patients (20%). Median survival was 10 months in all patients, 11.4 months in the wild-type K-RAS group, and 3.3 months in the mutant K-RAS group (P = 0.056). Point mutations in plasma DNA were identical to mutations found in the tumors, confirming the tumor as the source. In two patients with K-RAS mutations pretreatment, posttreatment plasmas were evaluated: a patient with clinical progressive disease retained the mutant DNA, while in a patient with a complete response (CR), the K-RAS mutation was no longer detectable. In an ongoing analysis of S0003 patients, to date, K-RAS mutations have been found in the plasma of 14 patients (10.1%). In 42 matched pre- and posttreatment specimens from both phase I and S0003 trials, four patients had K-RAS mutations pretreatment that were not detectable posttreatment. Of these, two had a clinical CR or partial response, and the other two had stable disease. It was concluded that detection of tumor DNA in plasma is feasible using molecular techniques and that this approach shows promise for monitoring patient response to therapy.