For automatic fetal head measurements and brain abnormalities diagnoses，the detection of fetal skull automatically is needed.So，a method for fetal head skull ellipse detection in ultrasound image is proposed.The main process is as follow: firstly the K-mean algorithm is used to cluster pixels into 3 classes and the bright objects are identified； then only the bigger connected components in the bright objects are preserved and skeletonized； finally randomized Hough transform (RHT) is applied to skeletonized components for ellipse detection.The adopted preprocessing method greatly reduces the number of pixels to be examined by RHT and improves the detection speed.Furthermore，a new scoring mechanism for RHT is introduced，which takes into account the head curves performance in both image and parameter spaces and increases detection accuracy.