In this Letter, the authors propose a novel convolutional neural network (CNN)-based estimation of the distance between an ultra-wideband (UWB) transmitter and receiver for a localisation. By exploiting the UWB signal characteristics, such as high-resolution in the time domain, the CNN is designed. The proposed CNN-based method estimates the distance from only the received signals, without signal-to-noise ratio information which is used for the conventional time of arrival (TOA)-based methods. Furthermore, as verified by simulation, the proposed CNN-based method significantly outperforms the conventional TOA-based method with respect to the distance estimation accuracy.