Precise fundus image features detection is an important step for diabetic retinopathy screening. Noises in fundus image features extraction need to be reduced. Based on Phase Congruency (PC), a new fundus blood vessel detection method is presented. Using PC model, feature detection is performed respectively on the original and contrast enhanced green component of color fundus image. To obtain sharp blood vessels features, logical AND operation is applied on the two PC feature detection results. Thresholding and morphological operation are performed so that blood vessels are visible. Finally, eyeball edges are removed to get blood vessels only. Experiment results show its effectiveness. It achieves expected effect that main blood vessels are clear while tiny vessels are complete as well.