Although predictive human performance modeling has been researched for 30 years in HCI, to our knowledge modeling has been conducted as a solitary task of one modeler or, occasionally, two modelers working in tight face-to-face collaboration. In contrast, we used predictive human performance modeling in a collaborative, distributed mode and reflect on that experience. We discovered that our tool for modeling, CogTool, while sufficiently functional and expressive to perform the modeling task, did not support collaborative, distributed modeling as well as we would like. We suggest process improvements in model construction, the management of assumptions, consistency, and communication, and suggest design solutions for the future of CogTool or other modeling tools. We further speculate on the generalization of our experiences to other types of usability evaluation when conducted in a distributed, collaborative environment. © 2012 ACM.