A model for negotiation was proposed which could self-adapt to interests of opposite side under the incomplete information. The expression of programs of interests to both sides was defined, the most bad weights algorithm of eliminating round after round was proposed, and the weights information was optimized, which was mostly close to the actual weights of the other side. On the basis of this model, a "hill climbing"-search algorithm was put forward to solve negotiation models with multi-attribution based on the traditional game theory field. The experimental results showed this model has good symmetry and fairness. The negotiation results are close to the Parreto-optimality.