The aim of this paper was to investigate 
the performance of simple convolutional neural network. 
This network includes four layers. 
The first layer and the third layer 
are feature detectors whereas the second layer is a feature pooling layer. 
The last layer of the network act as linear classifier. 
The network was evaluated on the MNIST database. 
The on-line backpropagation algorithm was used to train this network. 
A character distortion method applied to increase the diversity of the training data. 
Experiments have shown this neural network 
can yield performance comparable to the state-of-the-art on handwritten digit recognition.