The application and use of Compact Thermal Models of single chip electronic packages in a board level environment has seen limited study. In particular the influence of board thermal characteristics on the predictive accuracy of CTMs generated independently of the board is not well understood. A systematic study of the influence of board conduction on the predictive accuracy of Compact Thermal Models of BGA and CPGA package styles was performed. For the CPGA special attention was given to the socket model. The compact model parameters were extracted using a standard optimization procedure to best reproduce the junction temperature and heat flows computed for the detailed models exposed to a reduced set of heat transfer coefficients on the prime lumped areas. Various resistance network topologies were assessed for each package style. Detailed (FE) isotropic board models with conductivities spanning three orders of magnitude were created to test the influence of board conductivity on CTM accuracy. The board models included fully detailed isotropic models of the lSOP and 1S2P JEDEC standard thermal test boards. Compact models were deployed on these detailed board models with the board surfaces exposed to a wide variety of convective cooling conditions to test the limits of the package models. Results were compared to equivalent fully-detailed (FE) package models on detailed board models. The predictive capability of the optimized topologies was strongly correlated to the existence of strong local temperature gradients in the board, whenever the sensitivity to board heat conduction was high. When the board is poorly conducting (k=l W/m-K), board heat conduction is too low to matter. When the board is highly conducting (k=lOO W/m-K), the high conductivity smooths out local board temperature gradients, and the CTMs are accurate even though the board heat flow path is dominant.For intermediate board thermal conductivities of order 10 W/m-K, the heat conduction into the board is significant enough to cause sensitivity, and the conductivity is low enough to allow significant board temperature gradients. These conditions significantly degrade the predictive accuracy of CTMs.In general, optimized network topologies that included shunt resistances were found to predict the junction temperature to within 5% to 8% over a wide range of board conductivities, while star-shaped networks were generally only accurate to 10-15%. detailed model, printed circuit board, network topology