Research on facial expression recognition for establishing intelligent human-machine interfaces has been actively conducted. In recognizing human facial expressions, it is essential to learn new patterns of facial expression that appear with the lapse of time. In this paper, we propose a facial expression recognition model incorporating adaptive resonance theory in a counter propagation network, and evaluate the additional learning function of the model when targeting three types of facial expressions. In addition, we propose a brightness value correction processing method for reducing influence of difference in illuminance, and evaluate its usefulness.