A　new　effective　and　fast　minimization　approach　is　proposed　for　the　prediction　of　protein　folding,　in　which　the　＂relative　entropy＂　is　used　as　minimization　function.　Unlike　the　energy　minimization　method,　the　essence　of　this　approach　is　to　search　the　conformation　with　high　occupation　probability,　which　corresponds　to　the　state　with　low　free　energy　instead　of　low　energy.　The　off-lattice　model　is　used,　and　the　prediction　just　focuses　on　the　frame　of　the　main　chain　of　protein.　In　this　approach,　only　the　distances　between　the　consecutive　C-alpha　atoms　along　the　peptide　chain　and　a　generalized　form　of　the　contact　potential　for　20　types　of　amino　acids　are　used.　Tests　of　the　prediction　algorithm　are　performed　on　real　proteins　with　the　initial　structure　fully　denatured.　The　root　mean　square　deviations　(RMSD)　of　the　structures　of　four　folded　target　proteins　versus　the　native　structures　are　from　5　to　7　Angstrom.　The　advantage　of　this　approach　is　its　simple　potential　function　and　fast　performance.　Moreover,　it　can　be　considered　as　an　improvement　on　the　energy　minimization　method　in　principle.