The UK Meteorological office introduced a new bogussing scheme for tropical cyclone (TC) motion prediction in its global model in October 1994 after a series of tests. Before the operational implementation of this scheme, a 19-day period (in August and September 1994) was selected during which both the new and the old schemes were run operationally in parallel for a total of 12 TCs in three ocean basins. Comparisons between the forecast errors of the schemes demonstrated the superiority of the new scheme at all forecast time periods (Heming et al., 1995). This paper presents the results of an investigation into the possible reasons for the improved predictions by the new scheme during this trial period. All three factors of the new scheme (an improved initial positioning of the TC, an imposed steering flow derived from the past 6-h motion of the TC and an empirical vortex structure determined from actual observations and TC warning bulletins) are found to contribute towards the improvement. In particular, a good initial position together with an appropriate steering produce a much better 24-h forecast. For the 48- and 72-h forecasts, the steering flow remains an important factor and the contribution of the vortex structure also becomes significant in some cases. The steering flow appears to be the most crucial factor in determining the accuracy of the model predictions.