A grey particle filter (GPF) that incorporates the grey prediction algorithm and wavelet transform into the particle filter (PF) is presented. The GPF self-estimates the depth of maneuvering autonomous underwater vehicle (AUV) using the data measured by sensors equipped in the AUV under the condition that the prior maneuvering information is unknown and the measurement noise is time-varying. To implement the proposed method, the particles were sampled by standard sampling and grey prediction to insure the particles contain enough information about the true state of the maneuvering AUV. In addition, the measurement noise covariance was modified by wavelet transform in real time. Therefore, the GPF can effectively correct the prior distribution and likelihood function of the particles and then alleviate the sample degeneracy problem which is common in the particle filter. A high accuracy depth trajectory, which tracks by the outside position sensor as the true depth of the maneuvering AUV, was employed. Then the performance of the EKF, MMPF and GPF was evaluated through the experimental data. The results show the effectiveness and robustness of the GPF.