In this paper a new set of rotation invariant features for image recognition is introduced. The features are the magnitudes of complex Wavelet Transform (WT) of the image. The proposed method offers bigger advantages over Zernike Moment (ZM), for example, the Hamming Distance (HD) between the feature vectors of the different classes are bigger because WT can extract the corresponding local features in the different areas. The performance of the method is experimentally tested on a 26-class data set involving differently oriented binary images. The set consists of 624 images of all English characters. Using Hamming network 99.7 % and 98% classification accuracies are obtained respectively by WT and ZM.