Empathy, the ability to share the other's feeling, is one of the effective elements in promoting mutual reliability and construction of a good relationship. In order to create empathy between human-robot, a robot must be able to estimate the emotion of human and reflect the same emotion on its expression. In general, emotion can be estimated based on observable expressions such as facial expression, or unobservable expressions such as biological signals. Although there are many methods for measuring emotion from both facial expression and biological signals, few studies have been done on the comparison of estimated emotion. In this paper, we investigate whether emotion estimated from facial expression or biological signals could lead to empathy toward a robot. Using our proposed emotion estimation system, we performed two experiments and found that higher impression was rated on sociability elements with significant when the reflected emotion is estimated from uncontrollable emotion.