This paper proposes a Meta-learning and Markov Chain Monte Carlo (MLMC) based approach for blind single image super-resolution (SISR). The method learns kernel priors from organized randomness without requiring pre-training or labeled data. The approach consists of two phases: a Markov Chain Monte Carlo Kernel Approximation (MCKA) phase and a meta-learning based alternating optimization (MLAO) phase. In the MCKA phase, random Gaussian distributions are used to generate rational kernel priors, which are then used to optimize the kernel approximation network. In the MLAO phase, a meta-learning based optimization strategy is used to refine the blur kernel and restore the high-resolution image. The MCKA phase provides organized randomness to prevent the optimization process from getting stuck in local optima, while the MLAO phase ensures better convergence performance. The proposed method achieves superior performance and generalization ability compared to state-of-the-art methods on both synthetic and real-world datasets. The code is available at https://github.com/XYLGroup/MLMC.This paper proposes a Meta-learning and Markov Chain Monte Carlo (MLMC) based approach for blind single image super-resolution (SISR). The method learns kernel priors from organized randomness without requiring pre-training or labeled data. The approach consists of two phases: a Markov Chain Monte Carlo Kernel Approximation (MCKA) phase and a meta-learning based alternating optimization (MLAO) phase. In the MCKA phase, random Gaussian distributions are used to generate rational kernel priors, which are then used to optimize the kernel approximation network. In the MLAO phase, a meta-learning based optimization strategy is used to refine the blur kernel and restore the high-resolution image. The MCKA phase provides organized randomness to prevent the optimization process from getting stuck in local optima, while the MLAO phase ensures better convergence performance. The proposed method achieves superior performance and generalization ability compared to state-of-the-art methods on both synthetic and real-world datasets. The code is available at https://github.com/XYLGroup/MLMC.