This paper presents a method for ship detection using texture statistics from optical satellite images. The proposed method focuses on the extraction of ship candidates. First, a structural texture descriptor derived from local multiple patterns is introduced to describe image texture features, and then two statistical histograms are generated by quantizing texture features to describe the texture difference between sea and ships. Second, corresponding confidence maps representing the probabilities of ship candidates are created based on back projection of the statistical histograms, and ship candidates are extracted according to the confidence maps. Finally, the prior knowledge of ship shapes is employed to remove the false ship candidates. As using texture features, the proposed method is insensitive to different waves, illumination changes, ships with different sizes and bright/dark intensities. Experimental results demonstrate the method has good performance in both precision and recall.