Chemical exchange saturation transfer (CEST) MRI offers a powerful molecular imaging tool by exploiting the proton exchanges between free water and labile exchangeable molecules in vivo. Accurate CEST quantification from Z-spectrum analysis, particularly at clinical field strengths, suffers from low contrast-to-noise ratio (CNR). In this study, we proposed the use of non-local means (NLM) for CEST-MRI denoising to improve the CEST contrast quantification. The experimental results showed that the NLM denoising effectively removed noise presented in the low signal-to-noise ratio (SNR) CEST images with high resolution and well preserved the tissue structures. The CEST contrast map generated from the NLM-denoised images was notably improved, which was closer to the map generated from images with 3.5-fold higher SNR and meanwhile had much less partial volume effect. NLM denoising holds great potentials for the clinical CEST-MRI quantification improvement without compromising spatial resolution.