This paper proposes a variance estimator for the OLS estimator and nonlinear estimators such as logit, probit, and GMM. The estimator enables cluster-robust inference when there is two-way or multi-way non-nested clustering. It extends the standard cluster-robust variance estimator for one-way clustering and relies on similar weak distributional assumptions. The method is easily implemented in statistical packages like Stata and SAS. The paper demonstrates the method through a Monte Carlo analysis of a two-way random effects model, a placebo law example, and applications to empirical studies with two-way clustering. The paper discusses the challenges of clustering in empirical research, where errors may be correlated within clusters. It introduces a less parametric approach to cluster-robust inference that generalizes one-way cluster-robust standard errors to non-nested multi-way clustering. The method is useful in various applications, including cross-sectional studies with multiple levels of clustering, discrete regressors, pairwise observations, and panel data. The paper also addresses the importance of controlling for clustering in sample design and panel survey data. The paper presents a variance estimator for the OLS estimator and extends it to m-estimators and GMM estimators. The estimator is based on a weighted double-sum over all observations and is similar to spatial HAC estimators. The paper discusses practical considerations, including small-sample corrections and computational challenges with multi-way clustering. It also addresses the issue of positive-definite variance matrices and provides methods for handling non-positive-definite matrices. The paper concludes with applications to various econometric models and highlights the importance of cluster-robust inference in empirical research.This paper proposes a variance estimator for the OLS estimator and nonlinear estimators such as logit, probit, and GMM. The estimator enables cluster-robust inference when there is two-way or multi-way non-nested clustering. It extends the standard cluster-robust variance estimator for one-way clustering and relies on similar weak distributional assumptions. The method is easily implemented in statistical packages like Stata and SAS. The paper demonstrates the method through a Monte Carlo analysis of a two-way random effects model, a placebo law example, and applications to empirical studies with two-way clustering. The paper discusses the challenges of clustering in empirical research, where errors may be correlated within clusters. It introduces a less parametric approach to cluster-robust inference that generalizes one-way cluster-robust standard errors to non-nested multi-way clustering. The method is useful in various applications, including cross-sectional studies with multiple levels of clustering, discrete regressors, pairwise observations, and panel data. The paper also addresses the importance of controlling for clustering in sample design and panel survey data. The paper presents a variance estimator for the OLS estimator and extends it to m-estimators and GMM estimators. The estimator is based on a weighted double-sum over all observations and is similar to spatial HAC estimators. The paper discusses practical considerations, including small-sample corrections and computational challenges with multi-way clustering. It also addresses the issue of positive-definite variance matrices and provides methods for handling non-positive-definite matrices. The paper concludes with applications to various econometric models and highlights the importance of cluster-robust inference in empirical research.