The Community Atmosphere Model version 3 (CAM3) temperature simulation bias is examined in this paper. We compare CAM3 output with European Centre for Medium-Range Weather Forecasts (ECMWF) 40 year reanalysis (ERA-40) data. We formulate a time mean temperature bias equation then evaluate each term in the equation. Our focus is on the Northern Hemisphere winter time. We group the temperature equation terms into these categories: linear advection terms, nonlinear advection terms, transient eddy terms and diabatic heating, and find that linear advection and diabatic bias are the largest. The nonlinear terms (velocity bias advection of temperature bias) are much smaller than each of the other groups of terms at all levels except near the surface. Linear advection terms have dipolar pattern in the Atlantic (negative NW of positive) which reflects the shift of the CAM3 model North Atlantic storm track (NAST) into Europe, especially in the upper troposphere; opposite sign dipolar structure occurs over Alaska (positive) and the north Pacific storm track (negative). The transient advection terms in middle latitudes are larger in the upper troposphere and generally positive along the Atlantic storm track. Along the north Pacific storm track (NPST), the transient terms are negative in the mid and lower troposphere over much of the NPST (positive in upper troposphere). The diabatic heating bias has large values in the tropics along the Intertropical Convergence Zone (ICZ) and along the midlatitude storm tracks. During this time of year the ICZ is mainly in the Southern Hemisphere, but CAM3 emphasizes an ICZ-like heating in the northern hemisphere of the Atlantic and Pacific Oceans. CAM3 tends to have a weaker ICZ, especially in the Atlantic. In midlatitudes, we find large bias in heating by precipitation and vertically averaged net radiation over the NAST, Europe, and the Middle East.