CFD analysis is a powerful tool to obtain detailed data on the flow structure and heat transfer in turbine cascades. However, a systematic work has to be performed in order to validate CFD models including selection of a turbulence model, specifying adequate inlet conditions, evaluation of the grid dependence, etc. At that, among other data of practical interest, the local heat transfer is most sensitive to peculiarities of secondary flows and, consequently, to details of physical and computational modelling. Experimental data obtained by Langston et al. [1977], Graziani et al. [1980], and Holley et al. [2006] for flows in a large-scale turbine blade cascade are selected for validation of CFD results in the present contribution. Similar CFD efforts were made previously by many authors (e.g., Ameri et al. [1994], Lee et al. [1997], Ivanov et al. [2002], Levchenya et al. [2006, 2007]) using different computational grids, numerical schemes, and turbulence models. Generally, the experience accumulated evidences that, a practically grid-independent pressure and velocity distribution can be obtained with computational grids of about half million cells and a solver of second order spatial discretization. However, in order to get grid-independent data on the local heat transfer one needs much finer grids. Moreover, the solution obtained can be affected by the turbulence model, mainly because the multi-vortex 3D flow structure at the junction between the endwall and the blade leading edge is especially sensitive to the level of the eddy viscosity generated by the model.