This study was performed to assess the ability of the average linewidths of the methyl and methylene resonances from the proton nuclear magnetic resonance (NMR) spectrum of human plasma to distinguish reliably between a normal, apparently healthy population and untreated patients with cancer as was suggested by a recent report (Fossel et al, N Engl J Med 1986; 315: 1369-1376). In that report, the absence of overlap between the two populations suggested that the technique had great diagnostic potential. Blood samples were collected from healthy hospital personnel and patients with biopsy proven cancer who had not yet undergone chemotherapy or radiotherapy. NMR measurements were obtained on decanted plasma samples. In addition, to determine whether the distributions obtained were sensitive to experimental conditions, we varied a number of experimental parameters. These were storage temperature, observation temperature, NMR observation technique, and magnetic field inhomogeneity. In our hands, the distributions substantially overlapped, although the means of the average linewidths of the normal (35.0 +/- 5.3 Hz, range = 24 to 48 Hz) and patient (31.6 +/- 5.7 Hz, range 20 to 44 Hz) populations were significantly different, delta = 3.4 Hz, p = 0.023. The degree of overlap corresponded to 74 percent sensitivity and 59 percent specificity. The distributions were independent of the experimental conditions except for observation temperature. In this case, there was insufficient difference between our experimental conditions and those of Fossel et al (25 degrees C versus 22 degrees C) to explain the difference in results. Our findings suggest that wide variations exist among normal, apparently healthy persons and emphasize the need for carefully matched control subjects. We conclude that this measurement is, as of now, unsuitable as a method for identifying malignant disease.