Travel time and travel cost are key variables for explaining travel behaviour and deriving the value of time. However, a general problem in transport modelling is that these variables are subject to measurement errors in transport network models. In this paper we show how to assess the magnitude of the measurement errors in travel time and travel cost by latent variables, in a large-scale travel demand model. The case study for Stockholm commuters shows that assuming multiplicative measurement errors for travel time and cost result in a better fit than additive ones, and that parameter estimates of the choice model are impacted by some of the key modelling assumptions. Moreover, our results suggest that measurement errors in our dataset are larger for the travel cost than for the travel time, and that measurement errors are larger in self-reported travel time than software-calculated travel time for car-driver and car-passenger, and of similar magnitude for public transport. Among self-reported travel times, car-passenger has the largest errors, followed by car-driver and public transport, and for the software-calculated times, public transport exhibits larger errors than car. These errors, if not corrected, lead to biases in measures derived from the models, such as elasticities and values of travel time.