As the number of music and users in music streaming services increases, the role of music recommender systems is getting important to make it easier for users to find music that matches their tastes. The collaborative filtering paradigm is the most commonly used paradigm in developing recommender systems. Many studies have proven that deep learning is able to improve the performance of matrix factorization. One such method in deep learning that has been adapted for use in Recommender Systems is Autorec, which is a variation of the Autoencoder technique. Autorec shows that it performs better than the baseline matrix factorization using Movielens and Netflix datasets. Therefore, in this study we propose the use of Autorec to develop a recommender system for music. The experimental results show that Autorec performs better than Singular Value Decomposition (SVD), with an RMSE difference of 0.7.