Abstract. This paper studies the properties of a helpful and trustworthy explanation in a movie recommender system. It discuss the results of an experiment based on a natural language explanation prototype. The explanations were varied according to three factors: degree of personalization, polarity and expression of unknown movie features. Personalized explanations were not found to be significantly more Effective than nonpersonalized, or baseline explanations. Rather, explanations in all three conditions performed surprisingly well. We also found that participants evaluated the explanations themselves most highly in the personalized, feature-based condition.