Automated negotiation mechanisms can be helpful in contexts where users want to reach mutually satisfactory agreements about issues of shared interest, especially for complex problems with many interdependent issues. A variety of automated negotiation mechanisms have been proposed in the literature. The effectiveness of those mechanisms, however, may depend on the characteristics of the underlying negotiation problem (e.g. on the complexity of participant’s utility functions, as well as the degree of conflict between participants). While one mechanism may be a good choice for a negotiation problem, it may be a poor choice for another. In this paper, we pursue the problem of selecting the most effective negotiation mechanism given a particular problem by (1) defining a set of scenario metrics to capture the relevant features of negotiation problems, (2) evaluating the performance of a range of negotiation mechanisms on a diverse test suite of negotiation scenarios, (3) applying machine learning techniques to identify which mechanisms work best with which scenarios, and (4) demonstrating that using these classification rules for mechanism selection enables significantly better negotiation performance than any single mechanism alone.