Ultrasound can be perceived through bone conduction by the profoundly deaf as well as by normal-hearing subjects. Moreover, speech signals modulated onto ultrasound can be detected through bone conduction. This study explored how well listeners can understand ultrasonic speech and the confusion patterns to evaluate and improve bone-conducted ultrasonic hearing. The intelligibility of Japanese words classified by familiarity and Japanese monosyllables with bone-conducted ultrasound was investigated. Results showed that the intelligibility of familiar words was higher than that of unfamiliar words. Further, the results of a monosyllable intelligibility test with bone-conducted ultrasound and those of a test with air-conducted sound showed a similar pattern of speech recognition with regard to the errors made. The relationship between speech intelligibility and sound level showed that the increase in the intelligibility of bone-conducted ultrasonic speech did not exceed the increase in the intelligibility of air-conducted speech as the sound level rose.