We propose a fully differentiable architecture for learning an interpretable rule list classifier.