Extracting relevant feature and classification are significant in brain-computer interface (BCI) systems. Deep learning have achieved remarkable growth in many fields like speech recognition and computer vision. However, deep learning in biomedical field is yet to be fully utilized. In this paper, We propose a novel methodology for convolutional neural network (CNN) based motor imagery (MI) classification using new form of input. Continuous Wavelet Transform (CWT) is applied to the input Electroencephalography (EEG) signal to extract the features of MI. After transformation, we consider the real part and imaginary part of the transformed signal to exploit magnitude and phase information at the same time. This feature is fed to the CNN having one convolution layer, one max-pooling layer and one fully connected layer. The classification accuracy is tested on two public BCI datasets: BCI competition IV dataset IIb and BCI competition II dataset III. The proposed method shows increase in classification accuracy compared to other MI classification methods. The results show that the method using CNN with magnitude and phase based features can be better than other state-of-the-art approaches.