A three-dimensional robot vision sensor based on both active triangulation and a method using astigmatism is described. Range data obtained through active triangulation have comparatively high accuracy. However, range cannot be determined uniquely from a single-input image. On the other hand, range data obtained by the method using astigmatism are determined uniquely, but one cannot expect a high degree of accuracy. To improve the performance of the sensor, a novel method that fuses the range data obtained by these methods is proposed. Measurement results of the pilot system and verification of the performance of the system are presented.