Central pattern generator (CPG) is a kind of neural circuit which can be observed in many animals showing rhythmic patterns of actions. The CPG neural circuit can produce complex rhythmic patterns by receiving only simple signals from the brain. Generally, the CPG neural models can be applied to solve robotic problems, or to understand the underlying neural mechanism for rhythmic animal behaviours. In this paper, we focus on how a small number of neurons generate the variable frequencies and phase of motor actions, and inspect what is the capacity of a varying number of neurons as a CPG model. The performance measure consists of frequency variability, input/output response rate, and phase shift. We have used evolutionary computation to measure the best performance for each number of CPG neurons ranging from two to eight neurons, and the result shows that four neurons or more can easily generate variable frequencies and anti-phase difference for left and right motor actions.