Bearings are the most sensitive part of a motor drive system and predicting their health condition is very crucial in estimating the system’s Remaining Useful Life (RUL). Precise estimation of RUL is essential to increase machine reliability, decrease fault incidents, and prevent unscheduled downtime. The frequency, amplitude, and patterns of the vibration signals generated as a result of mechanical movements within the bearing system are indicative of its health status and the developments in the field of sensor technology have fueled data-driven approaches to discover the degradation trends by the analysis of condition monitoring data. Deep learning serves as a powerful and valuable prediction tool that can utilize raw data directly, eliminating the necessity for feature extraction. Spectrogram depicts the variation of frequency content in a signal over time and can serve as a potential diagnostic tool for detecting bearing irregularities. In this work, the spectrograms derived from vibration signals have been used to train a convolutional Neural Network (CNN) to predict the RUL of a bearing. The experimental data collected from the PRONOSTIA setup has been used for this purpose. The CNN model has been trained and tested and the results are presented. The performance of the model has been compared with similar works in the literature and found to be good.