We propose an improved neural-network-based palm biometric system with a rotation detection mechanism. An efficient method is developed for detecting the palm rotation angle in the scanned image by establishing reference line segments and calculating the segment slopes. By integrating this mechanism into our previous work, the limitation of palm rotation angle can be alleviated. Besides, the feature point searching step in can be simplified. Experimental results have shown that the rotation of palms in scanned images can be detected and the feature points can be located correctly.