Siamese neural networks (SNNs) are an essential sub-part of the deep learning family. They are great data interpreters which compare the feature representations of the two input instances to categorize them as similar or dissimilar. Siamese nets work great in situations where data availability is limited or the class imbalance problem prevails, thus outperforming the classical models. The work focuses on understanding the loss functions, space metrics, sub-network types, and performance metrics that built a SNN architecture. The paper reviews more than 30 research works with an objective of highlighting the power of Siamese networks in improving the overall results and their appropriateness for working in data-scarce environment. The application area and the dataset characteristics are essential factors in building the SNN’s architecture. Also, it is observed that the selection of loss functions and structural configurations significantly impacts the efficiency of SNN models. This paper will be valuable to researchers in various fields, providing insights into essential concepts and aiding in the development of task-specific models.