A cubature Kalman filter (CKF) phase noise compensation algorithm of the block sub-symbol second-order Lagrangian interpolation (BSSLI) is proposed to reduce the serious influence caused by phase noise in the coherent optical orthogonal frequency division multiplexing (CO-OFDM) system. The algorithm first uses the phase noise value of the time-domain OFDM symbol estimated by the pilot to perform the first second-order Lagrangian interpolation (SLI) and obtain a rough estimation of the phase noise, and then divides each OFDM symbol into several sub-symbols in the time domain. The estimated value of each sub-symbol is used as the estimated value of its intermediate sampling point to perform the second SLI, thus both the time resolution and the compensation accuracy are improved, and finally the CKF is used to compensate the residual phase noise. The simulation results show that: for 16 order orthogonal amplitude modulation (16QAM) and the laser linewidth of 800 kHz, the bit error rate (BER) of the algorithm can be close to 10^-7 and the error floor is reduced; for 32 order orthogonal amplitude modulation (32QAM) and the laser linewidth of 800 kHz, the BER of the algorithm can reach below 10^-3. Furthermore, compared with the linearly interpolated sub-symbol optical phase noise compensation scheme (LI-SCPEC) algorithm and the Lagrange interpolation based extended Kalman filter (LRI-EKF) algorithm, the proposed algorithm can better compensate the phase noise. Therefore, it can effectively improve the performance of the CO-OFDM system.