A novel analytical system was developed to trap and preconcentrate bismuth from the vapour phase stream. Bismuthine formed by sodium tetrahydroborate reduction was trapped on a tungsten coil previously heated to 270 degreesC. The analyte species were re-volatilised by increasing the coil temperature to 1200 degreesC and then transported to an externally heated silica T-tube by using a mixture of argon and hydrogen as the carrier gas. The base width of the transient signal was less than 0.5 s. The reproducibility of the analytical signal was influenced by the rate of heating of the tungsten coil at the releasing stage. The precision of the analytical system was found to be 5.8% RSD (n = 13) for 0.010 ng ml(-1) Bi when peak height was used. Due to sharp signals, peak area readings gave higher RSD values but they could be used for quantitation. The relationship between the analytical signal and the sample volume was found to be linear between 0.5 ml and 60 ml when using 0.100 ng ml(-1) Bi. When the sample volume was increased to 25 ml, the blank values became significant. The calibration plot for an 18 ml sampling volume was linear between 0.030 and 0.500 ng ml(-1) Bi. A limit of detection of 0.0027 ng ml(-1) (3s) was obtained with 18 ml of sample volume. The concentration limit of detection achieved was comparable with or better than those obtained by techniques such as ICP-MS and HG-ETAAS. When the distance between the trap and atomiser was increased to 200 cm, a 72% and 24% decrease was observed in the peak height and peak area, respectively. In order to validate the accuracy of the method, two geological standard reference materials and one certified water standard reference material were analysed and the results were found to be in good agreement with the certified values at the 95% confidence level.