Since variations in exogenous factors could cause unpredictable fluctuations/variability of the speed–flow relationship, the aim of this study was to develop probabilistic speed–flow relationships in a work zone (i.e. a segment of road in which maintenance or construction operations affect the operational characteristics of traffic flow). With observed traffic data from six highway work zones in Shanghai, China, lognormally distributed speed–flow functions were established to formulise probabilistic speed–flow relationships under uncongested and congested traffic conditions. A work zone capacity distribution model was then derived based on these probabilistic speed–flow relationships. The modelling results showed that work zone length, percentage of heavy vehicles and road roughness had strong effects on the probabilistic speed–flow relationships in work zones. A smaller mean work zone capacity was strongly associated with a long work zone length, a large proportion of heavy vehicles and poor road conditions. The models developed in this study could help traffic engineers evaluate the variability of work zone traffic states and the reliability of traffic management strategies.