In recent years, denoising diffusion models have demonstrated remarkable performance in the realm of generative modeling.However, these models have a significant drawback which is that of slow sampling time. The main cause of their high sampling cost is due to the Gaussian assumption of the reverse Markov transition, which requires a large number of denoising steps. To address this issue, we present a novel method, called DiffusionCNF, which models the reverse process of the diffusion model using normalizing flows. Our proposed approach enhances the sampling speed while maintaining the desirable properties of diffusion models. By leveraging the strengths of both models, we contribute to advancing the field of generative modeling and offer a promising solution for efficient and effective generation of complex data distributions.