To enhance the quality of Advanced Process Control (APC), the optimization of the sampling plan in critical dimension (CD) metrology is studied through empirical considerations concerning the characteristics of errors and a statistical approach. The metric of the optimization is the accuracy of lot mean estimation. Critical dimension errors are classified into static and dynamic errors. A static error is defined as an error that repeats through lots while keeping its tendency, and a dynamic error as an error whose tendency changes by lot. In the basic concept of our sampling plan, sampling positions and size are determined from the static error and dynamic error, respectively. The balance of the sampling number of the wafer, field, and site under the restriction of total sampling size is optimized by a statistical theory.