Margin in the context of SVMs is the distance between the decision boundary (hyperplane) and the nearest data points. Maximizing this margin often leads to better generalization. A larger margin implies a more confident and robust classification boundary.