A linear correlation between the depth and the length of the weld pool is found in laser beam welding experiments with varied laser beam power and constant welding speed. On the other hand, the weld pool length changes only slightly with increased welding speed and constant laser beam power. The existing analytical and numerical models fail to explain these dependences. The observed effects are essentially conditioned by the fluid flow in the weld pool caused by the thermocapillary effect, by the friction forces of the metal vapour passing through the capillary and by the convexity of weld pool and fusion zone caused by thermal expansion of the weld pool and the joined workpieces. In order to predict the weld pool length more accurately the model developed by Sudnik et al in 1996 is enlarged by the heat transport produced by the recirculating flow in radial sections of the weld pool. Verification of the model for 16MnCr5 steel with sheet thicknesses of 2 and 6 mm shows that it is suitable for predicting the weld pool geometry and for analysing the thermodynamics of the process. In order to gain a better understanding of the structure of heat transport in the weld pool, the different modes of transport are compared in respect of their contribution to the depth-to-length ratio of the weld pool. A calculation of the weld pool length for welding speeds of 1-8 m min −1 with a laser beam power of 2.5 kW shows that the relative contributions of the transport modes are as follows. Approximately 50-90% of the weld pool length (increasing with welding speed) results from conductive and translatory heat transport (with the fusion zone convexity contributing approximately 20-30%). The remaining 50-10% of the weld pool length (decreasing with welding speed) result from convective heat transport. The model predicts the shoulder in the weld pool trough. It also explains the change in the weld pool length by the effect of the gap width, by the transition from through welding to penetration welding and by improvements in beam quality.