The aim of signature recognition is to recognize name of author of the signature. It is a difficult work because scanned images are used for signature recognition. Identifying signatory is behavioral biometrics and has lot of potential of research in this area. There is a scope of accuracy improvement because of high variations of signature of same user. This paper investigates how can best feature extracted for signature recognition. In this research, pre-trained CNNs were used to fetch features and then fetched features were entered into machine learning algorithm, such as Random Forest, KNN, NN, and Decision Tree. This research is a comparative study on performance of model for particular datasets.