This study was performed to systematically evaluate the variability in tumor response assessments that occurs depending on how many tumor deposits are selected for measurement at imaging. The two largest perpendicular diameters of all tumor deposits in 36 patients were measured on computed tomography scans obtained at baseline and first posttherapy follow-up. A computerized modeling analysis of those data was performed to determine each patient's therapeutic response category assignment for every possible number of lesions in a grouping. The variance in the sum of measurements of these lesion groupings was calculated, and the frequency of response assessment categories was plotted against the number of lesions. The computerized analysis of the resultant 1,833,821 possible combinations of tumor deposits showed that when six lesions were measured bidimensionally and four lesions were measured undimensionally, the average variance decreased by 90%. The number of different response assessment categories into which a patient was assigned decreased with increasing lesion grouping size. When six or more lesions were measured bidimensionally, 9% of all possible lesion groupings still fell into a second response category, reflecting the effect of which particular lesions are chosen for measurement. Measuring larger numbers of lesions will decrease the variance. In this population, the variance decreased by at least 90% when six or more lesions were measured bidimensionally. Further confirmatory studies with larger series of patients are warranted before adopting this number as a criterion in clinical trials for assessing the activity of antineoplastic therapies.