Convolutional Neural Network (CNN) is the latest method in object recognition which is the development of Multilayer Percepton (MLP). CNN has a special layer of convolution layer and layer max pooling to process feature training hierarchically from data, with spatial data inputting. As a method of handwriting recognition method, CNN method is the best compared to method or similar model at this time. Therefore the CNN method is used in this research, for the recognition of Lontara script handwriting character. Using the CNN M8 model and TensorFlow open-source library for Lontara script image data training. After that made simple android interface as a medium of input Lontara script write hand through canvas for performance evaluation. From the test results obtained accuracy of up to 90%, by testing as many as 30 forms of Lontara.