Human behaviour must be described on the four levels known as policy, strategy, tactic, and operation. From the upper level (policy) to the lower (operation), the corresponding decision-making process goes through two successive contextualisations. At the first level, decision making has a fixed part (policy and strategy) and a dynamic part (tactic and operation), at the second level. Real-time aspects of decision making are on the dynamic part, i.e., in strong connection with the context in which an actor makes a real-time decision. We propose a modelling of real-time decision making by a contextual methodology that is composed of a series of 10 steps crossing the four levels. The aim is to support the self-training of actors to modify weaknesses of their behaviour during task realisation. The contextual methodology here proposes a unified framework on aspects that are generally contrasted – procedure versus practice, task versus activity, logic of functioning versus logic of use – and explains the two types of contextualisation in a context-oriented model with three layers. The contextual methodology has been applied in a road-safety domain (in which the car driver is continuously in a situation of real-time decision making) and is now used in another application in Open Source domain.