Traditional methods for detecting symmetry in image suffer greatly from the contrast of image and noise, and they all require some preprocessing. This paper presents a new method of image symmetry detection. This method detects symmetry with phase information utilizing logGabor wavelets, because phase information is stable and significant, while symmetric points produce patterns easy to be recognised and confirmable in local phase. Phase method does not require any preprocessing, and its result is accurate or invariant to contrast, rotation and illumination conditions. This method can detect mirror symmetry, rotating symmetry and curve symmetry at one time. Results of experiment show that, compared with pivotal element algorithm based on intensity information, phase method is more accurate and robust.