Fashion plays a pivotal role in society. Combining garments appropriately is essential for people to communicate their personality and style. Also different events require outfits to be thoroughly chosen to comply with underlying social clothing rules. Therefore, combining garments appropriately might not be trivial. The fashion industry has turned this into a massive source of income, relying on complex recommendation systems to retrieve and suggest appropriate clothing items for customers. To perform better recommendations, personalized suggestions can be performed, taking into account user preferences or purchase histories. In this paper, we propose a garment recommendation system to pair different clothing items, namely tops and bottoms, exploiting a Memory Augmented Neural Network (MANN). By training a memory writing controller, we are able to store a non-redundant subset of samples, which is then used to retrieve a ranked list of suitable bottoms to complement a given top. In particular, we aim at retrieving a variety of modalities in which a certain garment can be combined. To refine our recommendations, we then include user preferences via Matrix Factorization. We experiment on IQON3000, a dataset collected from an online fashion community, reporting state of the art results.