Identifying regions of loss-of-heterozygosity (LOH) in a tumor sample is a challenging problem. State-of-the-art computational approaches can infer LOH from single-nucleotide polymorphism (SNP) array data, but calling precise boundaries is complicated by normal-cell contamination and markers that are homozygous in the germline and therefore non-informative. More recently, the focus has shifted to pinpointing the loci recurrently affected by LOH events across multiple tumors. Recurrent LOH regions often harbor genes important for tumor suppression. Here, we propose a method that infers LOH rates across an entire sample set on an SNP-by-SNP basis. Our method achieves this by leveraging the straightforward principle that, by definition, LOH depletes heterozygotes, thereby disrupting Hardy–Weinberg equilibrium. We apply a statistical test for such LOH-influenced disruptions, and derive a maximum-likelihood estimator for the LOH rate based on the observed number of heterozygotes. This accounts for LOH in both its hemizygous deletion and copy-neutral forms, and does not make use of matched normal genotypes. Power simulations show high levels of sensitivity for the statistical test, and application to a control normal-tissue data set demonstrates a low false-discovery rate. We apply the method to three large publicly available tumor SNP array data sets, where it is able to localize tumor-suppressor gene targets of the LOH events. Inferred LOH rates are quite concordant across platforms/laboratories and between cell lines and tumors, but in a tumor type-dependent fashion. Finally, we produce rate estimates that are generally higher than previously published, and provide evidence that the latter are likely underestimates.