To examine the limitations of reverse transcription (RT)-PCR for the detection of circulating tumor cells in blood, we established a RT-PCR for carcinoembryonic antigen (CEA). Whole blood (10(7) nucleated cells) was mixed with cells from the colon cancer cell line LS174T (concentrations ranging from 0 to 10(6) cells). RT-PCR was performed to detect CEA mRNA in blood under various conditions. Healthy blood donors (n = 24) were examined by the established method for detecting CEA mRNA in blood. We were able to show that there is a detection limit for RT-PCR of 10 tumor cells in total and of 1 tumor cell in 10(5) nucleated cells. To obtain these results, a high number of PCR cycles (first PCR, 30 cycles; nested PCR, 45 cycles) was required. Under these PCR conditions, we found a positive PCR signal in 33% of healthy blood donors (n = 8). To overcome this problem, we reduced the nested PCR to 35 cycles. At that point, none of the controls showed a positive signal for CEA, and there was a subsequent decrease of the detection limit to 1 tumor cell in 10(2)-10(3) nucleated cells, lower than the detection limit of an immunocytological examination (1 tumor cell in 10(4) nucleated cells). When the amplification was performed with the tumor cells only and with no nucleated blood cells present, under exactly the same conditions, there was still a detection limit of 1 tumor cell in 106 nucleated cells. Our data clearly show that there is a severe loss of expected sensitivity of RT-PCR if it is performed in blood or nucleated blood cells. We conclude that PCR for CEA mRNA expression is not more sensitive than immunocytology and is, furthermore, plagued by the problem of a high percentage of false positive results.