Before the on-line processing in done, the system needs to be trained using a set of training samples. A Back Propagation Artificial Neural Network was used to perform our classification task. The developer of the Artificial Neural Network has to decide which configuration of the Artificial Neural Network is appropriate for a good out-of-sample prediction. The configuration of the Artificial Neural Network needs to be determined accurately to give an optimal classification result. In our experiment, this configuration included three layers. The first layer is the input layer consisting of 22 neurons, which is the number of features used to describe an image of wheat leaf. The third layer is the output layer and holds only one neuron. While training, the output is set to 1 if the training-image is that of a diseased leaf, else it is set to 0. After conduction experiments on all possible values from 1 to the size of a feature-vector, setting the number of neurons in the hidden second layer to 4 neurons gives the highest possible success-rate in our case.