In bilateral negotiation procedures, there often exist a number of issues. Models based on agenda, similarity or cases, ignore in most cases the interdependence of values of each issue. This paper proposes an interdependence-oriented bilateral multi-issue negotiation model. Firstly, the model adopts the thoughts of sequential procedure and local acceptance strategy for multi-issues negotiation. Secondly, it introduces off-line learning mechanism to partition the successful historical records of learning and uses these association rules and neural networks, which are generated from off-line learning to predict the acceptable values of interdependent issues. Lastly, this model presents a segmentation time strategy, which is based on the interdependent predicted value. The experiment results have shown that this model can improve the overall utility and efficiency to some extent.