Metal–organic frameworks (MOFs) present a combinatorial design challenge. The structural building blocks of MOFs can be combined to synthesize a nearly infinite number of materials. This suggests that computational tools, rather than experimental trial and error, can be used for high-throughput screening. Here, in the context of methane storage, we report the first large-scale, quantitative structure–property relationship (QSPR) analysis of MOFs. We investigated the effect of geometrical features, such as pore size and void fraction, on the simulated methane storage capacities of 130 000 hypothetical MOFs at 1, 35, and 100 bar at 298 K. From these data we developed models that can predict methane storage with high accuracy, based only on knowledge of the geometric features. Several models were developed: multilinear regression (MLR) models, decision trees (DTs), and nonlinear support vector machines (SVMs). In each case, 10 000 MOF structures were used to “train” the QSPR regression models, and the accuracy of the predictions was evaluated on a test set of 120 000 MOFs. The nonlinear SVM models can predict the methane storage capacity of MOFs in the test set with R2 values of 0.82 and 0.93 at 35 and 100 bar, respectively. Decision tree models produced rules for optimal design: for methane storage at 35 bar, MOFs should have densities greater than 0.43 g/cm3 and void fractions greater than 0.52; for methane storage at 100 bar, MOFs should have densities greater than 0.33 g/cm3 and void fractions greater than 0.62. Using two-dimensional response-surface analyses of the SVM models, we developed new hypotheses about combinations of material properties, yet unexplored, that might lead to very high methane storage capacities and warrant further investigation. SVM-based predictions of methane storage from MOF structural features can be tested online at our Web site: http://titan.chem.uottawa.ca/woolab/MOFIA.