One of the most challenging biometric authentication problems in recent years that we experience in daily life is signature verification. Signature verification systems are classified into two main approaches: offline systems and online systems. The offline signature verification systems are more difficult than the online systems since online systems have further information, such as velocity of writing, motion style and pen pressure, which allow for extracting of more features. This paper presents a deep learning method based on using the convolutional neural network (CNN) model for solving the offline signature verification problem to prevent the process of faking signatures that thieves practice. The CNN model was applied for extracting features and classifying whether the signature is genuine or forged. Our proposed method succeeded in achieving an accuracy of 94.73% on CEDAR dataset by using two types of signatures: genuine signatures and skilled forged signatures to test the performance of the system, which indicates that the method was effective and it can be supported by more feature extractors to get better results.