To assess the accuracy of large-eddy simulation (LES) predictions for flow without and with heat transfer in a rod bundle, analyses were performed with a constant-coefficient Smagorinsky LES model, and numerical predictions were compared with experimental measurements in a heated triangular rod array. First, flow simulations without heat transfer were performed with one and two channels at the central region of the bundle, and simulation predictions were compared with the experimental data. For the normalized mean axial velocity and the axial component of the turbulent intensity, the predictions of the one-channel model are nearly identical with those of the two-channel model. For the other turbulence parameters, the predictions of the one-channel model are either identical or are mostly in good agreement with those of the two-channel model. LES predictions for the mean axial velocity agree well with experimental measurements. Predictions of the axial component of the turbulent intensity agree well with experimental measurements for most of the points of measurement. Predictions of the other parameters of turbulence agree well to reasonably well with measurements. Because LES simulations are computationally very demanding, the LES simulation of heat transfer was performed only with the one-channel model. LES predicts the temperature of the rod surface within the range of the experimental error. The profile (log law) of the dimensionless fluid temperature T+ predicted by LES has the same slope as that derived from the measurements, but it has a significantly higher constant. The turbulent intensity of temperature is predicted well to reasonably well. The turbulent heat flux in the axial direction and the radial direction is predicted well at points away from the wall, but there is significant discrepancy between predictions and measurements close to the wall. The predicted turbulent heat flux in the azimuthal direction agrees very well to quite well with measurements.