The adaptive recursive least square （RLS） method was introduced for the first time to suppress the low-frequency （0 ~ 500 Hz） background noise of 100 km transmission optical fiber hydrophone system， and compared with the direct subtraction （DS）. The results showed that DS had poor basic noise suppression effect for fiber-optic hydrophone system， and RLS adaptive algorithm had obvious noise suppression effect： 16 dB@40 Hz， 20 dB@100 Hz， 15 dB@200 Hz ， 18 dB@300 Hz and 13 dB @400 Hz. After RLS adaptive noise suppression， the noise level of the system above 350 Hz was stable at -95 dB， which was equivalent to that of the high frequency part. RLS adaptive cancellation could suppress the noise and retain the useful signal in the sensor probe.