Multi-label classification is a supervised learning task where each data item can be associated with multiple labels simultaneously. Although multi-label classification models seem powerful in terms of prediction accuracy, they have however like mono-label classifiers certain limitations mainly related to their opacity. We propose in this preliminary work a novel approach for explaining multi-label classification models based on formal concept analysis (FCA). The proposed approach makes it possible to answer certain questions that a user may ask such as: What are the minimum attribute sets allowing the classifier f to make a prediction ? and What are the attributes that contribute to a given prediction?