Character recognition has been an active area of pattern recognition, not only because it improves man machine communication, but also because it provides a solution for processing large volumes of data automatically. Many systems of characters [e.g. Latin, Chinese, Korean, etc.] have been investigated for automatic recognition using both off-line (i.e. digitized images) and on-line data acquisition techniques. The purpose of this work is to recognize different forms of printed arabic characters written in three different fonts(Times new roman, Arial and Tahoma) using back-propagation neural network. The character recognition process must be proceeded by many necessary process to improve and prepare the character image to be ready for feature extraction. The processes include removing the noise if any , apply image binarization by using global threshold and then specifying the boundaries of the character in the image which is followed by another binarization process using local threshold. The character is then thinned using a two- step thinning process, the preprocessed character is input to feature extraction stage which determine the features of the character out of 100 different features which where defined for the whole set of arabic characters and the character features are used in three layer backpropagation neural network for character classification. This work was tested on a sample of printed characters and the correct average recognition rate was 97%.