Traffic congestion is growing in major cities, and, consequently, delays are becoming more frequent. Route guidance systems can significantly reduce delays by assisting drivers in finding alternative routes. Due to simplicity and scalability, the linear predictors have been an essential part of route guidance systems in predicting the future travel times. This paper investigates the limitations of linear predictors, and proposes a switching model which can predict travel times with less error. Real world data is used to evaluate the proposed switching predictor and compare the results with commonly used linear predictors.