Abstract: The ‘kissing-bond’ defect in friction stir welding (FSW) joints is difficult to detect with conventional ultrasonic method. A new ultrasonic feature imaging method was proposed for the problem. The method was composed of these steps: getting feature space of inspecting signal by decomposing time-domain signal with independent component analysis (ICA) method, finding out feature vectors for detecting targets according to energy transfer characteristic in ultrasonic detection, projecting ultrasonic signal to the feature vectors; obtaining feature parameter value by accumulating all the projections, using the feature parameter value to reconstruct the image of FSW joints. This method was used to conduct ultrasonic imaging for FSW joints with different tensile strength, and the images were also compared with ones generated by conventional C scan. The results showed that, inspecting results with this new method had good coherence with tension experiments, and showed a better detecting sensitivity for FSW joints evaluation than conventional C scan.