We have measured the supercurrent in aluminum atomic point contacts containing a small number of well characterized conduction channels. For most contacts, the measured supercurrent is adequately described by the opposite contributions of two thermally populated Andreev bound states per conduction channel. However, for contacts containing an almost perfectly transmitted channel 0.9 # t # 1 the measured supercurrent is higher than expected, a fact that we attribute to nonadiabatic transitions between bound states. PACS numbers: 73.40.Jn, 73.20.Dx, 74.50. + r In 1962, Josephson predicted that a surprisingly large supercurrent could flow between two weakly coupled superconducting electrodes when a phase difference d is applied across the whole structure. This phasedriven supercurrent I͑d͒ has subsequently been observed in a variety of weak coupling configurations such as thin insulating barriers, narrow diffusive wires, and ballistic point contacts between large electrodes. However, a theoretical framework powerful enough to predict the current-phase relation I͑d͒ in all configurations has emerged only during the last decade [1]. It applies in the mesoscopic regime, when electron transport between the electrodes is a quantum coherent process. Such transport is described by a set of N transmission coefficients ͕t i ͖ corresponding to N independent conduction channels. In the normal state, the conductance is given by G 0 P N i1 t i where G 0 2e 2 ͞h is the conductance quantum. In the superconducting state, electrons (holes) transmitted in one channel are Andreev reflected at the electrodes into holes (electrons) in the same channel. After a cycle involving two reflections at the electrodes, they acquire at the Fermi energy an overall phase factor p 1 d (Fig. 1). In a "short" coupling structure, these cycles give rise to two electron-hole resonances per channel, called Andreev bound states (AS) [2] with energies E 6 ͑d, t i ͒ 6D͓1 2 t i sin 2 ͑d͞2͔͒ 1͞2 (D is the energy gap in the electrodes). These two AS carry current in opposite directions, I 6 ͑d, t͒ w 21 0 dE 6 ͑d, t i ͒͞dd (where w 0 h͞2e), and the net supercurrent results from the imbalance of their populations. A quantitative comparison of the predictions of this "mesoscopic superconductivity" picture of the Josephson effect with experimental results is usually hindered by the fact that in most devices the current flows through a very large number of channels with unknown t i . However, an atomic-size constriction between two electrodes, referred to hereafter simply as an atomic contact [3], is an extreme type of weak coupling structure which accommodates just a few channels. Because their set ͕t i ͖ is amenable to a complete experimental determination and because it can be controlled in a certain range [4], atomic contacts are ideal systems on which to test quantitatively the concepts of mesoscopic physics. The knowledge of ͕t i ͖ allows in principle the calculation of all transport quantities. In particular, the phase-driven supercurrent is given bywhere n...