We introduce a new high-performance nonlocal filter, NESMA, for noise reduction in multispectral (MS) MR imaging. Through extensive analysis, we show that the NESMA filter demonstrates a high degree of overall image denoising while preserving edges and small structures. We compared the performance of the NESMA filter to the multispectral nonlocal maximum likelihood (MS-NLML) filter. Although the MS-NLML filter is highly efficient, it requires extensive computational time. NESMA markedly decreases computation time while maintaining comparable levels of noise reduction and feature preservation. Finally, we show that adaptive selection of similar voxels further improves filtering quality.