Taxis are a major means of public transportation in large cities, and speeding is a common problem among motor vehicles, including taxis. Unless caught by sensors or patrol officers, many speeding incidents go unnoticed, which pose potential threat to road safety. In this paper, we propose to detect speeding behaviors of individual taxis from taxi trajectory data. Such detection results are useful for driver risk analysis and road safety management. However, the taxi trajectory data are geographically sparse and the sample rate is low. Furthermore, existing methods mainly deal with the estimation of collective road speeds whereas we focus on the speeds of individual vehicles. As such, we propose to use a two-fold collective matrix factorization (CMF)-based model to estimate the individual vehicle speed. We have evaluated our method on real-world datasets, and the results show the effectiveness of our method in detecting taxi speeding behaviors. © 2018, Springer International Publishing AG, part of Springer Nature.