When designing and evaluating in-car user interfaces for drivers, it is essential to determine what effects these interfaces may have on driver behavior and performance. This paper describes a novel approach to predicting effects of in-car interfaces by modeling behavior in a cognitive architecture. A cognitive architecture is a theoretical frame-work for building computational models of cognition and performance. The proposed approach centers on integrating a user model for the interface with an existing driver model that accounts for basic aspects of driver behavior (e.g., steering and speed control). By running the integrated model and having it interact with the interface while driving, we can generate a priori predictions of the effects of interface use on driver performance. The paper illustrates the approach by comparing four representative dialing interfaces for an in-car, hands-free cellular phone. It also presents an empirical study that validates several of the qualitative and quantitative predictions of the model.