Empirical observations as well as theoretical analysis suggest that negotiation outcome in buyer-supplier situations can be improved by having accurate knowledge about the behavior of one's counterpart (i.e., negotiation partner). Yet, there is a paucity of research works dealing with the incorporation of learning methods into electronic procurement technologies, especially methods that can work with small amounts of information. This paper presents an application of nonlinear optimization for learning the parameters of a common negotiation decision function. Then, to show the usefulness of learning in a procurement-negotiation interaction, we outline a reaction algorithm that seeks to improve outcome. Detailed computational results with both the learning and reaction algorithms are conducted to demonstrate the viability of our approach.