Free-flow speed (FFS) is the speed of vehicles under low volume conditions, when the drivers tend to drive at their desired speed without being affected by control delay. Estimation of FFS is important in several applications. FFS varies extensively across various road facilities as they are influenced by driver behaviour, vehicle characteristics, road factors, landuse, geometric features, control factors, etc. The estimation of FFS in homogeneous traffic is comparatively simpler as the speed variation across vehicles is limited. However, in heterogeneous traffic conditions existing in countries such as India, the FFS distribution varies across vehicle classes. The studies conducted by the authors explored the FFS distribution of various vehicle classes such as two-wheelers, three-wheelers, cars, buses, etc. However, detailed analysis revealed that the variation in FFS can be better explained by further classification of vehicles into subclasses. The study also found that the vehicle's lane position is a factor affecting FFS. The study was conducted on four- and six-lane divided roads in Chennai, India. A total of 24 study sections were chosen for data collection. Speed data were collected during early morning hours to ensure free-flow conditions. The vehicle movements were recorded using video cameras. The details regarding site factors such as carriageway width, link length, landuse, presence of kerb and type of area were collected manually. The speed and lane data were extracted and tabulated from the video recordings. The authors studied the speeds of about 17,800 vehicles (36% two-wheelers, 8% three-wheelers, 8% buses, 33% cars, 10% light commercial vehicles and 5% trucks). The vehicles were classified into 14 subclasses and speeds were analysed. The study also evaluated the effect of lane position on FFS of different classes of vehicles. It was found that vehicles on kerb lanes experienced lower speeds than those on inner lanes. Furthermore, FFS models for four- and six-lane divided roads were developed using multiple linear regression. Significant difference in speeds was observed within and across subclasses of vehicles. The models also evaluated the effects of various road factors such as carriageway width, link length, adjacent landuse type and presence of kerb on FFS. Models such as these can find applications in planning and operational analysis of urban road facilities. © 2015 The Authors.