Efficiency measurement in public transport requires an adequate account of unobserved network characteristics that are typically modelled as factors separable from the production process. This paper proposes a panel data model that allows for non-separable firm-specific heterogeneity in an input distance function. The proposed model is applied to a sample of German and Swiss urban transit companies operating from 1991 to 2006. The results underline the presence of non-separable unobserved factors and their effects on technological characteristics such as returns to scale. Moreover, the data suggest that the effect of time-invariant heterogeneity could be significantly greater than technical inefficiency.