Convolutional Neural Networks (CNNs) have enabled significant improvements across a number of applications in computer vision such as object detection, face recognition and image classification. An audio signal can be visually represented as a spectrogram that captures the time-varying frequency content of the signal. This paper describes how a CNN can be applied to the spectrogram of an audio signal to distinguish pathological from healthy speech. We propose a CNN structure and implement it using Keras to test the approach. A classification accuracy of over 95% is obtained in experiments on two public pathological speech datasets.