Eighty-one recent carcinogenicity studies carried out by the National Toxicology Program (NTP) were evaluated to determine how the utilization of statistical analyses based on the proportion of animals with primary tumors (all sites) or the proportion of animals with malignant neoplasms (all sites) affected the interpretation of the data compared to analyses of site-specific effects. Utilizing site-specific analyses, the NTP concluded that 45 of the 81 studies (56%) showed carcinogenic responses, 7 (9%) produced equivocal effects, and 29 (36%) showed no evidence of carcinogenicity. An analysis of tumors at all sites often resulted in site-specific carcinogenic responses going undetected. Less than half of the 45 carcinogens identified as producing site-specific carcinogenic responses showed a significant increase in the incidence of primary tumors (22 chemicals) or malignant tumors (21 chemicals). Among the 29 chemicals interpreted as not carcinogenic based on site-specific effects, only two showed significant increases in overall tumor incidence. Two major problems are associated with an evaluation based on overall (all sites) tumor rates: The pooling of various tumor types reduces study sensitivity for detecting chemically related increases in site-specific tumor incidences, and the biological relevance of combining the incidences of tumors of varying morphologies and topographies is questionable. Most national and international guidelines for studying chemicals for carcinogenicity in rodents (or in humans) emphasize site-specific effects. Thus, despite purported advantages of analyses based on overall tumor rates (e.g., simplicity; reducing concerns regarding false positive results) primary emphasis should continue to be on site-specific analyses.