In this paper, an open source CFD code is used for predicting the mixed-convection flow of coolant in a low-voltage transformer winding. A significant feature of these flows is the high Prandtl number of the dielectric coolant fluid. It is found that for the geometry under consideration, the CFD model predicts the presence of hot-plumes in some of the horizontal cooling ducts. This appears to be due to the highly non-uniform mass flow distribution around the winding, which arises as a result of an inhomogeneous temperature profile across the individual vertical ducts (a direct consequence of high fluid Prandtl number). These hot-plumes are particularly problematic since local heating can cause thermal degradation of the paper insulation. It is also found that there is a strong coupling between the flow in different passes, which is communicated via persistent hot streaks in the fluid being convected from one pass to the next. The importance of these hot streaks is apparent from the CFD results, yet they are unresolved by the lumped parameter based ‘network models’ preferred by industry, which assume perfect mixing and homogeneous flow quantities in all cooling ducts. In particular, these methods are unable to account for the persistence of these inhomogeneous features from one pass to the next. While it is expected that CFD approaches should provide higher detail than network modelling, the insight gained from the present simulations would suggest that significant changes to current industrial approaches might be necessary in order to correctly account for the observed phenomena.