Although information extraction and coref- erence resolution appear together in many applications, most current systems perform them as independent steps. This paper describes an approach to integrated infer- ence for extraction and coreference based on conditionally-trained undirected graphical models. We discuss the advantages of condi- tional probability training, and of a corefer- ence model structure based on graph parti- tioning. On a data set of research paper cita- tions, we show significant reduction in error by using extraction uncertainty to improve coreference citation matching accuracy, and using coreference to improve the accuracy of the extracted fields.