Congestion on the Dutch motorway network is an actual problem, which originated over the past decades. Over the past few years, the extent of congestion is decreasing, though still very significant. Road authorities therefore show interest in traffic congestion forecasting. In this way they can inform road users or undertake other strategic actions. This thesis is therefore dedicated to the variability of traffic in congestion forecasting. The main objective is to develop a methodology operationalized in a model, which is able to predict congestion on motorways without knowledge of the actual traffic conditions. The model takes the variability of traffic into account and is substantiated with a solid theoretical framework relating the predictability of factors and their effects to traffic supply and demand. To fulfill this objective, a literature research has been conducted on traffic flow theory, factors and effects influencing traffic demand and traffic supply, model approaches and probabilistic methods. The identified influence factors are listed below: regular pattern of variation in human travel behavior over the day, over the days of the week and over the periods of the year; public holidays / vacation periods; events; weather conditions; road works; incidents; variations in vehicle population; variations in driver population; luminance; ‘intrinsic’ variations in driving behavior and in human travel behavior. Using the acquired knowledge a research methodology is developed. The model approach makes use of the basic principles of traffic flow theory based on the conservation of vehicles and first order traffic flow theory. To take the variability of influence factors into account, an intelligent sampling technique is used: Latin Hypercube Sampling. Before the model is constructed, the predictabilities and the effects of the various influence factors on traffic demand and traffic supply are described and explained through a theoretical framework. Some of them are always predictable (e.g. public holidays, luminance), while the predictability of others depend on data (e.g. road works or weather conditions). Incidents are considered not very predictable. The occurrence and therefore the effects of the identified influence factors can be continuously present or only on certain moments in time. They can also be on every cell of the motorway corridor or only on a selection of cells of the motorway corridor. The developed model makes use of traffic demand profiles and traffic supply variables. These are processed by a first order traffic model using a Godunov scheme. Traffic is numerically sent through the model subject to the defined boundary conditions. When the flow exceeds the capacity, congestion sets in and propagates backwards in space according to the first order traffic theory. From the modeled data, travel times and other performance indicators can be derived. A trajectory method is used to calculate the actual travel times. Before the model processes the traffic demand profile and traffic supply variables, these are corrected for the identified influence factors. The occurrence of these influence factors can be defined manually. However, to be able to incorporate the variability of traffic, a sampling component is added to the model. In this way the occurrence of the influence factors can be determined through a probability function. After multiple runs, the output indicators are collected. The model is calibrated and evaluated using data from the A27 motorway between Hooipolder and Gorinchem. The data was recorded over the year 2011. First the calibration of the traffic supply variables is performed. The calibration results showed a good likeness to the recorded travel times. Congestion in the model also sets in at similar moments in time and space as in the recorded data. Next the intrinsic variability of traffic is calibrated, the recorded travel time distribution over the year 2011 was very well approximated by the model. The research shows that the developed methodology is suitable for predicting travel times or other performance indicators. Producing traffic demand profiles is achievable, however the effects of some influence factors (road works, weather conditions) are not trivial. The estimation of the traffic supply variables from data showed to be a tougher task. The model results show relatively large uncertainties in the travel times as congestion was probable to set in (i.e. peak periods). When the traffic demand and traffic supply are close to each other, the probability on congestion increases. However, the difference in travel times when congestion sets in, compared to travel times where congestion remains absent, is relatively large. Hence, the large uncertainties in the peak periods. This implies that, even though the occurrence and effects of the identified influence factors are very accurately available, uncertainties in travel times can still be significant. Using the developed methodology, it is important to have reliable data sources regarding the identified influence factors. Especially for the influence factors that have significant effects on the traffic conditions (road works, adverse weather conditions, events), the availability of accurate data is highly needed. Predictions made with help of unreliable or incomplete data is deemed to be inaccurate. The case study results also lead to some recommendations. For operational use of the methodology, further analysis of the possibilities for extension of the model to larger networks (e.g. the Dutch motorway network) is recommended. The effects of the identified influence factors can then not be only adopted from theory. The prediction of the traffic conditions produced by the model shows less certainty as traffic demand and traffic supply values are close to each other, even though the influence factors are fully predictable and accounted for. Further research on this topic is also advised.