Working memory is an important brain function for memorizing information in everyday life. This study proposed an alternative approach for analyzing visual working memory based on brain complexity to identify brain state condition for memorizing new scenes. Multi-scale entropy (MSE) was used for analyzing the complexity of function Near Infrared Spectroscopy (fNIRS) data measuring the hemodynamic response of brain during a cognitive experiment. The results revealed the distinctive entropy between remembered and forgotten cases in premotor cortex area at FC3 position. The entropy of remembered case is higher than that of forgotten case; this could indicate that the brain requires more activity and, then, more hemodynamic responses for good memorizing. Hence, the results indicated the potential of using the complexity of fNIRS for characterizing the brain state for working memory.