Folding free-energy landscapes of a mini-protein, chignolin, were calculated by using multicanonical molecular dynamics simulations with each of four AMBER generalized Born methods to compare the accuracies of these methods. In simulations with the HCT, OBC I, and OBC II methods, the populations of the native structure were the largest and were more than 30%. In the GBn simulation, however, the population was the second largest and was only about 15%. The average errors of the OBC methods were smaller than that of the HCT method. We therefore concluded that the OBC methods are preferable to obtain a reliable free-energy landscape. (C) 2008 Elsevier B.V. All rights reserved.