A molecular dynamics (MD) method using the multicanonical algorithm, multicanonical MD, is proposed to enhance the efficiency of conformational sampling of peptides. Multicanonical MD is a constant temperature MD on a deformed potential energy surface, which gives a multicanonical ensemble characterized by a flat energy distribution. The multicanonical ensemble, in turn, is converted into a canonical ensemble by the reweighting formula of Ferrenberg and Swendsen. The multicanonical algorithm was originally developed by Berg et al. as a method of Monte Carlo simulations. The multicanonical MD is its application to MD simulations. Using a system of a simple double-well potential, it was confirmed that this method yields a correct canonical distribution. A test of the method was also performed with a five-residue peptide, Metenkephalin, and showed that multicanonical MD samples much larger conformational space than conventional canonical MD.