In this paper we extend the existing metrics for estimating difficulty, directed to action games (in particular, platform videogames) and taking into account different types of challenges. Specifically, we analyse how challenges are represented in time and space and observe the players' behaviour when facing such challenges. Accordingly, we propose a set of models to predict the probability of failure and success in different situations. This type of assessment can serve as a validation mechanism for automatic level generation algorithms and also to perform adaptive difficulty techniques. As a final point, we analyse multiple gameplay information retrieved from gaming sessions, in which 40 users performed over 10000 trials in different types of challenges. We have verified that the main hypotheses behind the proposed metrics are applicable and the estimators are valid approximations to the players' behaviour.