Experimental studies were carried out to evaluate the reproducibility in 3H measurements with presently available liquid scintillation counters. A set of 3H samples which have activities ranging from 10(4) to 10(7) dpm were prepared and measured repeatedly in every possible different order with six counters of different models. The results demonstrate that the measured counting rates of the samples are subjected to much greater variations than expected from counting statistics. The poor reproducibility was found to be attributed to rate dependent photomultiplier gain variations. The variations in counting rate as large as 10 per cent was observed in a particular counter. It is also found that rate dependent photomultiplier gain variations can cause the external channel ratio to shift, giving rise to error as large as 9% in the efficiency determination.