This paper gives advice on population size selection from literatures and analyzes the relations between population size, dimension, precision and stability. It proposes a method of population size selection. Two benchmark functions are employed to perform function optimization with different dimensions: 20, 60, 100 and 200. For each dimension, population size is set to 20, 40, 60 and 80. This paper analyzes the results and proposes a general population size. Furthermore, an approach based on experiments is proposed for population size selection. With the approach, a relation graph of population size, time and precision can be used to determine the optimum population size.