As urban roads expand over time, road surface features become more complicated and critical to driving safety. Constructing a database of road surface features through an automatic process is particularly crucial when these features are applied to the fields of facility maintenance, autonomous vehicle assistance, 3D cyber city construction, and navigation-information renewal. This paper proposes a fully automatic approach for reconstructing road surface features based on the terrestrial mobile light detection and ranging (lidar) technique. Analysis steps were developed and an experiment was performed using a real lidar data set. The results indicated that the proposed road surface reconstruction method is feasible and that the classification indices, correctness and completeness, were both higher than 90%. Finally, the uncertainty associated with each analysis stage was explicitly assessed through an error budget analysis. This can provide a guideline for practical applications and further system optimization.