The efficiency of detection of H- and K-ras mutations in 27 CD-1 mouse liver tumors by direct sequencing of polymerase chain reaction (PCR)-amplified DNA isolated from formalin-fixed and paraffin-embedded tissues was compared with that after assay by both NIH 3T3 transfection (followed by sequencing of amplified transformant DNA) and direct sequencing of PCR-amplified DNA isolated from frozen tumors. Some tumor samples were chosen for comparison because they contained ras mutations that were detected by either NIH 3T3 transfection or sequencing of PCR-amplified DNA derived from frozen tumors, but were not detected by both techniques. The efficiency of detecting K-ras mutations was similar for sequencing of amplified fragments derived from both paraffin-embedded tissues and from frozen tumors. However, these two techniques differed in their efficacy for detection of H-ras codon 61 mutations. Furthermore, this difference appeared to be mutation-specific: the sequencing of amplified products from paraffin-embedded tumor tissues allowed increased detection of CAA to AAA mutations but decreased detection of CAA to CTA mutations relative to sequencing of amplified fragments derived from frozen tumor DNA. Direct sequencing of PCR products from paraffin-embedded sections was more sensitive than NIH 3T3 transfection for detection of activated K-ras genes containing codon 13 mutations but less sensitive for detection of activated H-ras genes containing codon 61 mutations. In summary, direct sequencing of amplified DNA from either frozen tumors or formalin-fixed, paraffin-embedded tissues can be more sensitive than NIH 3T3 transfection for detection of codon 13-activated K-ras genes. However, it appears to be less sensitive than NIH 3T3 transfection for detection of certain activating H-ras mutations. Depending upon the questions being asked of the data, each of the methods can provide useful information about ras gene mutations in tumor samples. The apparent differences in sensitivities between the methods is not yet understood, but such differences should be considered in the analysis of data obtained when only one method is used.