Proteins are dynamical objects; their structure fluctuations are often the key to their function and keen attention need therefore be paid to it. Molecular dynamics (MD) simulations are a widely used technique to study dynamics of both proteins and nucleic acids. However, simulating molecular machines containing several thousands of amino acids, on long enough time scales to observe relevant structural deformation, remains challenging. Normal mode analysis (NMA) is better suited to study the slow dynamics of proteins. We briefly describe the theory underlying NMA and the simplifications used to render it tractable for (large) proteins. We also describe different kinds of analyses that can be performed on the eigenvectors to characterize the dynamical properties of the system. A number of validation studies are then summarized. Finally we describe NMA servers available on the Internet.