Non-negative matrix factorization (NMF) is a technique for finding parts-based, linear representations of non-negative data. While NMF often produces sparse representations, the degree of sparseness is not controllable. This paper introduces an extension of NMF that explicitly controls sparseness, leading to better parts-based representations. The method incorporates sparseness constraints into the NMF formulation, allowing for more direct control over the representation's properties. The paper also provides a complete MATLAB implementation of both standard NMF and the sparseness-constrained version. The algorithm uses a projected gradient descent approach, with a projection operator that enforces sparseness by controlling the $ L_1 $ and $ L_2 $ norms. The method is tested on various datasets, including face images and natural images, showing that sparseness constraints can lead to more meaningful and parts-based representations. The paper also discusses the relationship between the proposed method and other recent extensions of NMF, and concludes that the ability to explicitly control sparseness improves the effectiveness of NMF in data analysis tasks.Non-negative matrix factorization (NMF) is a technique for finding parts-based, linear representations of non-negative data. While NMF often produces sparse representations, the degree of sparseness is not controllable. This paper introduces an extension of NMF that explicitly controls sparseness, leading to better parts-based representations. The method incorporates sparseness constraints into the NMF formulation, allowing for more direct control over the representation's properties. The paper also provides a complete MATLAB implementation of both standard NMF and the sparseness-constrained version. The algorithm uses a projected gradient descent approach, with a projection operator that enforces sparseness by controlling the $ L_1 $ and $ L_2 $ norms. The method is tested on various datasets, including face images and natural images, showing that sparseness constraints can lead to more meaningful and parts-based representations. The paper also discusses the relationship between the proposed method and other recent extensions of NMF, and concludes that the ability to explicitly control sparseness improves the effectiveness of NMF in data analysis tasks.