This paper presents the lessons learned when a computational GOMS modeling tool was used to evaluate user interface concepts and team structure designs for a new class of military shipboard workstations. The lessons are both encouraging and cautionary: For example, computational GOMS models scaled well to a large and complex task involving teams of users. Interruptability and working memory constructs had to be added to conventional GOMS model concepts. However, two surprises emerged: First, the non-psychological aspects of the model construction were the practical bottleneck. Second, user testing data in this domain were difficult to collect and lacked definition, meaning that the model provided a better characterization of the design details than the user testing data. Included in these lessons are recommendations for future model applications and modeling methodology development.