The difficulty of phase unwrapping to recover the phase from noise,holes,fringe breaks,undersampling and other degrading influences is a great challenge. On one hand,least-squares approach may provide a smoothed phase,but it cannot limit the transfer of noise,as well as the smoothing characteristic causing errors. On the other hand,least-squares algorithm may fill the holes with data of approaching original phase. We aimed at improving the filling effect,limiting the transfer of noise,a new phase unwrapping algorithm based on binary mask and least-squares iteration is presented. One simulated result and two experimental examples of phase unwrapping using new method showed its validity,the obtained phase which approached original value very well and was much better than the results from single least-square unwrapping operation.