The prediction of critical heat flux (CHF) in rod bundles has been studied with both subchannel and bundle-average methods. The correlations of Biasi, Bowring, CISE-4, and Barnett were considered. The General Electric 9-rod bundle CHF data were used in the comparisons. Calculations were performed by the two-fluid subchannel code THERMIT-2. The results indicate that the subchannel method yields more conservative CHF predictions than the bundle average method. This is attributed to the two-phase turbulent mixing phenomenon in the bundle, which can be modeled only on a subchannel basis. The results also indicate that the CISE-4 correlation had the smallest error in prediction of transition boiling for both subchannel and bundle-average methods.