An　efficient　soft　docking　algorithm　is　described　for　predicting　the　mode　of　binding　between　an　antibody　and　its　antigen　based　on　the　three-dimensional　structures　of　the　molecules.　The　basic　tools　are　the　＂simplified　protein＂　model　and　the　docking　algorithm　of　Wodak　and　Janin.　The　side-chain　flexibility　of　Arg,　Lys,　Asp,　Glu,　and　Met　residues　on　the　protein　surface　is　taken　into　account.　A　combined　filtering　technique　is　used　to　select　candidate　binding　modes.　After　energy　minimization,　we　calculate　a　scoring　function,　which　includes　electrostatic　and　desolvation　energy　terms.　This　procedure　was　applied　to　targets　04,　05,　and　06　of　CAPRI,　which　are　complexes　of　three　different　camelid　antibody　VHH　variable　domains　with　pig　alpha-amylase.　For　target　06,　two　native-like　structures　with　a　root-mean-square　deviation　＜　4.0　A　relative　to　the　X-ray　structure　were　found　within　the　five　top　ranking　structures.　For　targets　04　and　05,　our　procedure　produced　models　where　more　than　half　of　the　antigen　residues　forming　the　epitope　were　correctly　predicted,　albeit　with　a　wrong　VHH　domain　orientation.　Thus,　our　soft　docking　algorithm　is　a　promising　tool　for　predicting　antibody-antigen　recognition.