A novel computational fluid dynamics analysis method of predicting semiconductor junction temperatures precisely without modeling printed circuit board (PCB) line patterns was developed. First, PCBs are divided into multiple regions. The effective anisotropic thermal conductivity of each region is then assigned as follows. All the regions are divided into smaller subregions whose size is below the pattern width. The thermal conductivity of each subregion is defined by the property of the material at the center of the subregion. Next, a thermal circuit network composed of all the subregions is generated, and finally the anisotropic thermal conductivities of each region are computed by solving this thermal network matrix. When boards are divided into multiple regions, there is a convergence region size under which the analytical results show no further change. In this paper, the relationship between the size of the divided regions and the accuracy of the analytical results was investigated. It was confirmed that the calculated semiconductor junction temperatures were precisely coincident with the experimental results when the size of the regions was less than 20 times the line pattern width.