Random forests are an effective machine learning model, this is why they are still widely used today. However, whilst it is quite easy to understand how a decision tree works, it is much more complex to interpret the decision made by a random forest, because it is typically the result of a majority vote among many trees. Here we examine various definitions of abductive explanations for random forests based on binary attributes. We are particularly interested in the generation problem (finding an explanation) as well as the minimization problem (finding a shortest explanation). We show in particular that irredundant abductive explanations (or sufficient reasons) can be difficult to obtain for random forests. We propose instead the notion of "majority reasons", that are in principle less concise abductive explanations, but which can be computed in polynomial time.