For medical applications of the in-line phase-contrast x-ray imaging, phase retrieval is a crucial step for quantitative imaging such as reconstructing the 3-D distribution of tissue linear attenuation coefficients and refraction indices. The conventional phase retrieval algorithms, such as the transport of intensity equation (TIE) based algorithms, are not robust against the quantum noise that appears in acquired images due to the radiation dose constraints in medical imaging. In this work a new attenuation-partition based iterative phase retrieval algorithm is proposed. This new algorithm takes advantage of the correlation between the attenuation and phase-shift, and is much robust against noise in acquired images. Phase retrieval results from experimental images show that this new iterative algorithm is fast and robust, and it has good potential for medical x-ray imaging applications. © 2008 Optical Society of America.