The variation of walking speeds for passengers passing through a facility within a congested underground station is known to be an important factor influencing pedestrian route choices. This paper describes a study of walking speed variation under unidirectional and bidirectional pedestrian flow conditions in Hong Kong, China, mass transit railway stations. The studied facilities selected for data collection and analysis include a unidirectional walkway and a bidirectional stairway. The variation in walking speed is due to increased passenger demand in the station when trains arrive. A study of the distribution of walking speed variation and relationships between average walking speed and variations of that speed was conducted based on data collected from an observation survey. For uncongested sections, when the mean walking speed decreased, the walking speed variation also decreased under both unidirectional and bidirectional pedestrian flow conditions. In addition, the walking speed variation was smallest when pedestrian flow approached capacity. However, in congested sections, when the mean walking speed decreased, the walking speed variation increased under both unidirectional and bidirectional pedestrian flow conditions. All walking time data were empirically proved to present a normal distribution. This paper presents newly developed walking speed variation models for three selected pedestrian facilities. These models can be further considered in connection with pedestrian simulation models to assess alternative improvement schemes at urban railway stations.