RNA arbitrarily primed (RAP)-PCR is a powerful tool for studying differential gene expression in cancer cells. Systematic analysis of human tumor samples may provide a list of markers with potential application to the diagnosis, prognostic assessment, and treatment of the disease. Nevertheless, because of characteristics inherent to the samples and technique, artifactual results are likely. We have assessed the effects of several factors on RAP-PCR performance to determine the sensitivity and reproducibility of the technique, as well as the accuracy of its results, under different conditions in human cell lines and in a series of 129 paired human normal colonic mucosa-colorectal carcinoma samples. Our results show that RAP-PCR provides reliable fingerprints in a relatively wide spectrum of circumstances, including variations in RNA concentration and contamination by DNA. Densitometric analysis indicated that relative band-intensity variations more than 20% were reproducible in 95% of the cases. Serial analysis of paired normal-tumor cases yielded a number of bands that were recurrently either underexpressed or overexpressed in tumor relative to normal mucosa. These differentially expressed bands are prime targets of research because they represent candidate tumor-specific up- or down-regulated genes with a relevant role in carcinogenesis.