In this paper we examine some assumptions commonly made in modeling call centers. In particular, we evaluate the assumption that agents are homogeneous, statistically equivalent servers. We examine empirical data to highlight the issues that create heterogeneity between agents. We explore a call center environment where agents increase their productivity over time, but eventually leave the organization. We consider the implication of this heterogeneity and explore a routing policy that attempts to exploit this heterogeneity and improve long-term call center performance. We consider the application of experience-based routing; that is, routing to agents based on their availability and experience relative to other available agents. We examine policies where calls are routed to the most experienced agents when the call center is busy, to facilitate efficiency, and to the least experienced agent when the call center is slow, to facilitate learning. We investigate the potential improvement in performance that can be achieved by considering agent experience when making routing decisions and characterize the conditions under which the improvement is most significant. We find that routing to agents based on experience can yield substantial improvements over a wide range of conditions.