This paper deals with the so-called `probability of scenarios' issues: what is the meaning of these numbers, and what is the appropriate degree of unlikelihood of extreme scenarios ? For far distant futures (such as assessments of climate change in 2100), scenarios without any quantified uncertainty level are often problematic, but forecasts with precise probabilities are out of reach. We propose to use a non-probabilistic approach: possibility theory. De Finetti's interpretation of uncertainty is used to define the meaning of quantified possibility levels based on acceptable betting odds. Turning to the question ``At which level of possibility should the possible futures be selected?'', we reason that a set of n scenarios should contain at least one future at possibility level 1, but if there are extremes cases their possibility should be precisely 1/n.