Technical modifications of the reverse-transcription/polymerase chain reaction (RT/PCR) amplification method now permit its use to detect amplified products from as few as one abnormal cell, either isolated or mixed with a larger number of normal cells. We studied the reproducibility of such results using as targets low numbers of cells from chronic myeloid leukaemia (CML) patients and CML cell lines in quintuplicate two-step RT/PCR designed to amplify BCR-ABL sequences. When one K562 or KYO1 cell was diluted in 10(3) non-CML HL60 cells, an amplification product was obtained in each test; at greater dilutions BCR-ABL transcripts were detected erratically. Titration of cDNA synthesised from 5 x 10(7) cells from four CML patients showed that whereas positive BCR-ABL sequences could be amplified in some tests starting with as little as a 1 in 10(7) dilution of cDNA template (corresponding to 5-10 cells), the dilution threshold for reproducible amplification was around 1 to 5 in 10(5) (100-500 cells). Quantitative PCR analysis revealed that reactions from 1 in 10(7) diluted cDNA contained less than 10 BCR-ABL transcripts as the starting template. The stochastic nature of the amplification from such small numbers of transcripts was illustrated by results of 10 replicate PCR tests on cDNA from a patient expressing both b3a2 and b2a2 transcripts: dilutions of cDNA up to 1 in 10(5) yielded dual transcript amplification in all 10 tests, but the 1 in 10(7) cDNA dilution resulted in b3a2 and b2a2 products in three tests, b3a2 only in three, b2a2 only in one and no amplification in three tests. We conclude that this 'sampling effect' may yield false-negative results and thus misinterpretation of data regarding assessment of gene expression when the quantity of target material available for study is very small.